NIOSH
EPA
     \ipev
United States
Department of Health
and Human Services
            Public Health Service
            Centers for Disease Control
            National Institute for Occupational Safety and Health
United States
Environmental Protection
Agency
            Office of Environmental
            Processes and Effects Research
            Washington, DC 20460
EPA-700/7-80-184
December 1980
            Research and Development
            A Prototype Gas
            Analysis  System
            Using a Miniature
            Gas Chromatograph


            Interagency
            Energy/Environment
            R&D Program
            Report

-------
        A PROTOTYPE GAS ANALYSIS SYSTEM USING

            A MINIATURE GAS CHROMATOGRAPH
  Professor James B. Angell, Principal Investigator
                    John  H.  Jerman
               Stephen C. Terry, Ph.D.
                    Soheil Saadat
            Integrated  Circuits  Laboratory
           Stanford Electronics Laboratory
                 Stanford University
             Stanford,  California  94305
               Contract  No.  210-77-0159
     U.S.  DEPARTMENT  OF  HEALTH  AND  HUMAN  SERVICES
                Public Health Service
             Centers  for Disease Control
National Institute for Occupational Safety and Health
    Division of Physical Sciences and Engineering
               Cincinnati, Ohio  45226
                     April  1981

-------
                         DISCLAIMER
Mention of company name or product does not
constitute endorsement by the National Institute
for Occupational Safety and Health.
             DHHS (NIOSH) Publication No. 81-115
                    Dr.  Laurence  J.  Doemeny
                    NIOSH  Project Officer
                              ii

-------
                                ABSTRACT
The techniques of integrated circuit processing have been utilized to
miniaturize the main components of a gas chromatography system.  A 1.5-m
long separating column, sample injection valve, and output thermal con-
ductivity detector are all fabricated on a single 5-cm diameter silicon
wafer.  These components are combined with a sophisticated microcomputer
system, consisting of a microprocessor, solid state memory, analog-to-
digital converter, keyboard, and display, to form an advanced, portable
gas analysis and data logging system.  The system can automatically
sample the air, measure the concentration of up to 10 contaminant vapors,
calculate and store contaminant concentrations, and estimate average
worker exposures.  The computer system allows exceptional flexibility in
the retrieval of concentration information, ease in operation, and sim-
plicity of calibration.  The instrument can be used as an exposure moni-
tor to sample the air at 1.5 minute intervals for an 8-hour shift, or
it can be used as a survey instrument to take and analyze samples and
provide concentration information within 30 seconds of sampling.  The
small size, advanced capabilities, and ease of operation of the system
should have a significant impact on the fields of industrial hygiene and
occupational safety and health.  The instrument will greatly expand the
ability to monitor workers' exposures to toxic gases and therefore aid
in the reduction of such exposures, and provide better epidemological
data in future health studies.
                                   iii

-------
                                CONTENTS

Abstract	iii
1.  INTRODUCTION 	   1
    1.1  Background	   1
2.  THE MINIATURE GAS CHROMATOGRAPH	   5
    2.1  Introduction to Gas Chromatography	   5
    2.2  Silicon Etching Technology	   6
         2.2.1  Etched Geometries	   6
    2.3  Miniature GC Fabrication	   9
         2.3.1  Capillary Column	   9
         2.3.2  Sample Injection Valve 	  10
         2.3.3  Detector Configuration and Fabrication 	  12
3.  INSTRUMENT SYSTEM
    3.1  Carrier Gas Supply	15
    3.2  Sample Injection System 	  16
    3.3  Analog Electronics	16
    3.4  Digital Electronics 	  19
    3.5  Package	20
4.  PERFORMANCE
    4.1  Overall Instrument Performance	23
         4.1.1  Analog Electronics Performance 	  28
    4.2  GC Column Performance	29
         4.2.1  Column Linigs	29
         4.2.2  GC Separations	31
         4.2.3  Optimization of the GC Column	34
         4.2.4  Temperature Dependence 	 ....  40
    4.3  Detector Performance	41
         4.3.1  Comparison with Other Detector 	  42
         4.3.2  Detector Operation 	  44
                                    v

-------
    4.4  Computer System	51
         4.4.1  Data Analysis	51
         4.4.2  Operating Program Flow	60
5.  CONCLUSION	65
REFERENCES	67
GLOSSARY	69
                                    vi

-------
                                FIGURES


1.1   Photograph of the prototype gas analysis system	3

2.1   Schematic of a simple gas chromatography system	5

2.2   The steps in the photolithographic process 	  7

2.3   Representative etched cross sections in Si 	  8

2.4   Photograph of the GC wafer	9

2.5   Cross-sectional view of sample injection valve . 	 11

2.6   SEM photograph of an etched silicon valve seat	11

2.7   Cross-sectional view of detector structure 	 12

3.1   Block diagram of prototype gas analysis system	15

3.2   Schematic of the carrier gas supply system	 . 16

3.3   Sample injection system	17

3.4   Schematic of detector current source and detector amplifiers . . 17

3.5   A/D converter and multiplexer system 	 18

3.6   Digital electronics block diagram	19

3.7   Layout of the prototype miniature gas analysis system	21

3.8   Photograph of the interior of the prototype instrument 	 22

4.1   Typical chromatogram from the reproducibility tests	25

4.2   System reproducibility vs sample size	26

4.3   Oscilloscope photograph of the amplified detector output
      or chromatogram	31

4.4   Miniature GC chromatogram using OV-101 stationary phase	32

4.5   Idealized chromatogram showing retention times and peak
      widths	33


                                   vii

-------
4.6   Golay plot of miniature chromatographic column with OV-101
      column lining	37

4.7   Optimum plate height and separation factor as functions of
      zo	38

4.8   Optimum carrier gas velocity as a function of Z	38

4.9   Number of effective plates as a function of maximum allowable
      carrier gas pressure and etched depth	39

4.10  Partition ratio as a function of 1/T for chloroform and
      methylene chloride 	 42

4.11  Detector cross section 	 ......... 44

4.12  Relative detector response for the three detector driving
      systems	45

4.13  Graphical solution to detector operating points for dif-
      ferent detector currents 	 47

4.14  Detector operating point temperature and nitrogen peak
      voltage as a function of detector current. .	49

4.15  Thermal conductivity of Ar/He gas mixtures as a function of
      the percentage of Ar in the mixture	50

4.16  Linearity of detector sensitivity as a function of argon
      mole fraction in the helium output stream	50

4.17  Impulse response, h(t), of the optimum filter	53

4.18  The resultant waveform after filtering a gaussian input
      signal	54

4.19  Graphical representation of the filtering process	56

4.20  Computer simulation of gaussian peak with noise	58

4.21  Flow chart for peak finding and area calculations	59

4.22  Simplified flow chart of the main program	60

4.23  Keyboard command description 	 . 	 62

4.24  Gas and system parameter listing 	 .......... 63
                                   viii

-------
                                 TABLES







4.1   Results of reproducibility test	24




4.2   Analog electronics performance 	  28
                                   ix

-------
                            1.  INTRODUCTION
1.1  BACKGROUND

The miniature gas chromatographic air analyzer is a unique instrument com-
bining a gas chromatograph fabricated on a silicon wafer and a sophisti-
cated microcomputer system for analyzing the data and controlling the
device.  The development of the miniature system has depended principally
on advances in the field of integrated circuit processing.  The reduction
in size of the gas chromatograph is due to the application of silicon
etching technology to the miniaturization of the separating column, injec-
tion valve, and detector.  The sophisticated electronics system needed to
control the system has been made possible through the use of recently
introduced one-chip microprocessors and high density, low power memory
chips.  The combination of these technologies results in an instrument
with unique analytical and field capabilities.

There is a widespread need in the industrial hygiene field for the rapid
and accurate monitoring of potentially hazardous concentrations of many
gases in the air.  Currently there are several types of portable, direct-
reading instruments for the monitoring of single contaminants.  These
instruments lack the versatility needed to adequately monitor the work
place.  A variety of small, passive monitors are becoming available, but
these either require expensive and time-consuming analysis in the labora-
tory or are designed for only a single contaminant.  A gas chromatograph
(GC), which is capable of separating, identifying, and measuring the con-
stituents of a gaseous sample, is an ideal instrument for performing the
analysis of an industrial environment.  A conventional  GC,  however, is
much to bulky to be used either for a portable survey instrument or as an
exposure monitor for a mobile worker.

The reduction in size of the components of a GC realized by using silicon
etching technology have made a portable instrument possible.  The feasi-
bility of such a portable instrument was studied at Stanford University
under NIOSH contract #210-76-0140 [l] and a prototype, which is the subject
of this report, has been fabricated.  The prototype is capable of automatic,
real-time analysis of atmospheric gas samples for up to 10 different vapors
at concentrations down to a few parts-per-million.  These instruments
promise to significantly aid in the investigation of potentially harmful
environments.

The work on the miniature GC started at Stanford University in 1971 in an
attempt to develop a miniature instrument for NASA to fly on the Viking
landers to Mars.  A separating column was rapidly developed, but there was
insufficient time to develop the integrated valve and detector structures

-------
before the launch.  The first integrated GC, which consists of a long
separating column, a sample injection valve, and a thermal conductivity
detector, all fabricated on a single silicon wafer, was developed in
1974 (Yj.  It operates by continually passing a carrier gas through the
separating column.  This column is lined with a material capable of ad-
sorbing and desorbing the constituents of a gaseous sample.  A brief
pulse of sample gas is injected at the head of the separating column.
As the constituents are swept along the column they are each retained in
the lining for characteristically different total times.  They thus exit
from the tube as separate peaks, which are detected by the changes in
thermal conductivity of the output gas stream.

The components of the integrated GC have been developed and refined in
the last few years and the performance of the device has steadily improved.
There have been concurrent rapid advances in the semiconductor industry
in the field of miniaturized digital computers.  A significant effort has
been expended to apply those miniature computers to the GC system to form
a portable analytical system with advanced capabilities.  The computer
controls the sequence of operation of the GC and implements sophisticated
digital filtering, peak detection, peak area calculations, data storage,
and display functions which would have been impossible in a portable
package the size of the prototype before 1979.

Work has also been done to miniaturize the other parts of a total instru-
ment, such as the carrier gas supply and sample injection system.  Since
the miniature GC is so much different from conventional instruments, most
available components are not suitable for use in this system.  The GC,
support components, and electronics system have been combined to form a
portable instrument, shown in Figure 1.1, about 20 cm x 22 cm x 10 cm
(8" x 9" x 4"), which weighs 2.8 kg (6.2 Ibs).  It is capable of sampling
the air automatically at specified time intervals, or on command from the
operator.  The instrument is programmed to look for up to 10 different
constituents of the gas sample, calculate their respective concentrations,
calculate and store time-weighted-average concentrations, display the
concentration of any desired compound on demand, and sound an alarm if
any concentration exceeds a preset level.  It contains sufficient carrier
gas in convenient, replaceable cartridges to last for at least four days
of continuous operation and has a rechargeable battery system for up to
8 hours of operation between charges.  Since the computer handles all
aspects of the instrument operation, the unit is very simple to use.

-------
Figure 1.1.   PHOTOGRAPH OF THE PROTOTYPE GAS ANALYSIS SYSTEM.   The package
             measures about 20 cm x 22 cm x 10 cm and weighs 2.8 kg.

-------
                   2.  THE MINIATURE GAS CHROMATOGRAPH
The miniature gas chromatograph is the heart of the portable gas analysis
system.  It separates the gaseous sample into its component vapors and
detects and measures the concentration of the separated constituents.
The miniature gas chromatograph is similar in function to larger labora-
tory chromatographs, but the method of fabrication of the components is
quite different.

2.1   INTRODUCTION TO GAS CHROMATOGRAPHY

Gas chromatography is a technique widely used in analytical chemistry for
the separation and analysis of gaseous samples.  A typical gas chromato-
graph  (GC) consists of a carrier gas supply, a sample injection system,
a separating column, and an output detector, as shown in the block diagram
(Figure 2.1).  Separation of the sample vapors is achieved via their dif-
ferential migration through a capillary column.  The sample vapors are
injected at the input of the column by a valve and are swept through it
by an inert carrier gas.  The column is lined with a liquid stationary
 CARRIER GAS
Us
y
TOR
f SAMPLE
GAS
,NSJAE^N CAPn±ARY T
VALVE COLUMN COf
D




HERMAL
»JDUCTIVITY
ETECTOR



DATA
PROCESSINC
UNIT
I
                                    FABRICATED ON SINGLE
                                    SILICON WAFER
       Figure  2.1.   SCHEMATIC  OF A SIMPLE  GAS  CHROMATOGRAPHY SYSTEM.
phase, a substance capable of absorbing and desorbing each of the com-
ponent vapors.  The migration rate of each vapor along the column depends
on the carrier gas velocity and the degree to which the vapor is absorbed
by the stationary phase.  The mixture of sample vapors, injected into the
capillary as a single pulse, is separated as it travels through the column,
with each component vapor traveling at a different rate and emerging at a

-------
specific time.  The column's output is thus a series of vapor peaks sep-
arated by regions of pure carrier gas.  To detect these peaks, the output
gas stream from the column is passed over a detector which measures a
particular property of the gas, such as thermal conductivity, which can
be related to the concentration of sample vapor in the carrier gas.  The
detector produces a signal which is amplified and recorded as a function
of time.  Chromatograms obtained in this manner are used in the quantita-
tive analysis of the sample mixture since the identity and quantity of
each vapor in the mixture can be determined from its retention time in
the column and the area under its output peak.

2.2  SILICON ETCHING TECHNOLOGY

Miniaturization of the components of the GC requires the ability to fabri-
cate small three-dimensional structures with tightly controlled geometries.
Photolithographic techniques have been developed by the integrated circuit
industry for patterning silicon with complex two-dimensional circuits with
features less than 5-ym wide.  The combination of these photolithographic
techniques with different silicon etches has resulted in the ability to
form the valves, columns, and detectors required for a miniature GC.

In larger gas chromatographs, the common practice is to fabricate the
individual elements of the system as separate units, and then assemble
the components into a system using connecting tubing.  In the integrated
GC, however, the volumes of the individual components (column, valve, and
detector) are so small that any discrete tubes and connectors would con-
tribute unacceptably large dead volumes which would excessively dilute
the sample peaks.  The miniature GC system must be connected with tubes
of diameter nearly equal to the column diameter, which necessitates the
integration of many of the system components on the same substrate.  With
silicon etching technology, the connecting tubes can be fabricated at the
same time and with similar dimensions as the system elements, so no extra
processing steps are required.

2.2.1  Etched Geometries

A wide variety of different features  can be etched in silicon depending
upon the shape of the mask and the type of etchant used.  Fabrication of
these features involves a series of oxidation, photolithography, and etch-
ing steps similar to those of  standard integrated circuit processing.  The
basic photolithography process is depicted in Figure 2.2.  A layer  of
silicon dioxide  (Si02) is  first grown on  the wafer  in a high temperature
furnace.  This oxide layer will  serve as the  etch mask in the subsequent
silicon etching  step.  The desired pattern must  then be transferred to the
oxide.  A thin layer of photosensitive material  is spun onto the wafer.
A glass photographic plate with  an emulsion  image of the desired pattern
is placed in  contact with the  wafer,  and the  photoresist is  exposed with
UV  light.   The photoresist  on  the wafer  is then  developed,  leaving the
oxide protected  by  the resist  in the  desired  etch mask pattern.  The
excess  Si02  is etched  away,  leaving bare silicon where the  etched  feature
is  to be  located.

Several cross-sectional profiles of grooves  are  available,  depending upon
the  orientation  of  the groove  with respect to the  crystallographic planes

-------
                                 PHOTOLITHOGRAPHY PROCESS
    Silicon Dioxide
       Photoresist

%
                               •Silicon
                        (a)
                                       (glass)
                                                              -Glass
Figure 2.2.  THE STEPS IN THE PHOTOLITHOGRAPHIC PROCESS ARE:   (a) a thin
             layer of S102 is thermally grown on the silicon surface;
             (b) a layer of photosensitive material is spun onto the oxide
             layer; (c) a glass mask is placed in contact with the wafer
             and exposed to UV light;  (d) the glass mask has a negative
             image of the desired pattern; (e) the unexposed photoresist
             is developed away; (f) the oxide layer is etched in buffered
             HF; and (g) the photoresist is chemically removed.
of the silicon, the type of etchant used, and the conditions under which
the etch is performed.  An anisotropic etchant such as potassium hydroxide
(KOH) in water will produce V-shaped grooves in  (100) oriented silicon,
as shown in Figure 2.3a.  This etchant will not  attack the  (111) crystal
planes in silicon, which form the V-walls.  For  a relatively narrow  groove,
where the V-walls meet at the bottom, the depth  of the grooves can be

-------
                (a)
              //////////I
                                            -Si02 MASK
                                     r//////////////////////
                                                      y/777///7/A
                                '36°!
                               SILICON
                (b)
                 (c)
     Figure 2.3.  REPRESENTATIVE ETCHED CROSS  SECTIONS  IN Si.
                  (a) Grooves etched with KOH  and water in (100)
                  Si.   (b) Grooves etched with KOH  and  water in
                  (110) Si.   (c) Grooves etched with HF-nitric
                  solutions in Si.
precisely determined by the initial width  of  the  etch mask.   In (110)
oriented silicon and along certain directions in  (100) silicon, a KOH
etch will produce grooves with precisely vertical walls as shown in
Figure 2.3b.  The choice of geometries  is  limited with these etches
because of their dependence upon the  crystal  structure of the silicon,
so that some desired features, such as  round  holes with vertical walls,
cannot be realized.

A mixture of hydrofluoric and nitric  acids can be used as an isotropic
etchant to produce grooves as shown in  Figure 2.3c.   This etchant is
used in forming many of the features  of the miniature GC system since it
produces nearly rectangular grooves,  oriented in  any direction of the
wafer.

-------
A sequence of these Isotropic and anisotropic etches is used to form the
substrate GC wafer.  Though described separately below, many of these
etches for the valve, capillary column, and detector gas channel can be
performed simultaneously, reducing the total number of necessary photo-
lithographies and etches.  The detector chip is removable from the sub-
strate wafer and is batch fabricated on separate wafers using similar
etching techniques.

2.3  MINIATURE GC FABRICATION

2.3.1  Capillary Column

The miniature capillary column is formed by etching a long, narrow spiral
groove into the surface of a silicon wafer and then hermetically sealing
the wafer to a Pyrex glass cover plate.  The capillary column is the
prominent feature in Figure 2.4, a photograph of the GC wafer.  Most of
the miniature GC's produced have used capillary columns 1.5-m long, about
200-ym wide, and 30-jjm to 55-ym deep.  Holes etched through the wafer at
the ends of the column serve as entrance and exit ports for the carrier
gas.
         Figure 2.4.   PHOTOGRAPH OF THE GC WAFER.  The view is
                      through the 5-cm square Pyrex cover glass.
                      (A) Carrier gas input feedthrough hole.
                      (B) Sample gas input and exit feedthrough
                      holes.   (C) Sample injection valve, back-
                      side of wafer.  (D) 1.5-m long separating
                      column.  (E) Detector gas channel, back-
                      side.   (F) Vent to atmosphere.

-------
The capillary column and gas feedthrough ports are fabricated in two
separate photolithographic and etching steps.  The starting material is
a 200-ym thick, 5-cm diameter silicon wafer of (100) surface orientation.
The capillary column is first formed by transferring the 1.5-m long
folded spiral into a 1-ym thick layer of Si02.  Since the spiral does
not follow any particular crystallographic direction, an isotropic HF-
nitric acid etchant is then used to etch the spiral groove into the
silicon.  The remaining SiC>2 is then removed, and more SiC>2 is grown as
the etch mask for the feedthrough hole etch.  The holes are etched through
the wafer using an anisotropic etchant in order to minimize their volume.

The wafer is again stripped of the remaining oxide and is placed in con-
tact with a 5-cm square piece of Pyrex glass.  A relatively low tempera-
ture anodic bonding process is used to hermetically seal the wafer to the
glass.  The silicon and glass are heated to about 400°C and a potential
of 600 V is applied to the structure.  The resulting electrostatic field
pulls the two materials together, and a bond is quickly formed.  The seal
between the glass and silicon is hermetic and irreversible, thus forming
the etched spiral groove into an enclosed capillary tube.  This unlined
capillary column must be coated with a column lining before any chromato-
graphic separations can be obtained.  The methods of column lining will
be discussed in Section 4.5.1.

2.3.2  Sample  Injection Valve

The purpose of the sample injection valve is to introduce a very brief
pulse of sample gas into the carrier gas stream at the head of the capil-
lary column.   For the miniature GC, the volume of injected sample is
about 10 nl, which generally requires that  the valve remain open for about
5 ms.  This volume of injected sample is about three orders of magnitude
smaller than what can be obtained with the  smallest commercial valves
and connecting tubing, so a custom, integrated one-valve injection system
has been developed for use with the GC column.

This injection system uses a miniature, normally  closed, solenoid-actuated
diaphragm valve with an etched silicon valve  seat,  as  shown in Figure  2.5.
The valve seat is  fabricated using two isotropic  silicon etches on the
opposite side  of  the wafer  from the  capillary column.   The details of  the
etched  valve seat  can be seen  in  the  SEM photograph, Figure 2.6.  The
nickel  diaphragm  is held against  the  valve  sealing  ring by the body  of
the solenoid.  An  adjustable compression spring normally forces the  sole-
noid plunger and  diaphragm  against the valve seat.  When the  solenoid  is
energized,  the plunger  is pulled  back into  the body of the solenoid, al-
lowing  the  diaphragm  to  relax.  The  gas on  the input  side  of  the  valve is
held  at a slightly higher pressure than the output,  so that  the gas  is
 forced  over the valve  seat,  injecting a pulse of  gas  into  the capillary
 column.  When  the  solenoid  is  turned off,  the plunger  again  forces  the
 diaphragm against  the  valve seat,  stopping  the flow of gas.   Since  the
valve  is fabricated directly on  top  of  the  capillary column,  there  is
 little  dead volume associated  with the  feedthrough hole connecting  the
valve  seat  and the column.   The  volume  of  the feedthrough  hole on the  out-
put  of  the  valve  is  only about 4  nl,  and  this volume is swept quickly  by
 the  2  to 5  ul/s  carrier gas flow.   Experiments with this valve design

                                     10

-------
                               SOLENOID
                               PLUNGER
               SOLENOID
                 BODY
                                       6-32  —»
                                   THREADED NUT
                                                     SEALING
                                                     RING
          NICKEL
  SILICON  DIAPHRAGM
SUBSTRATE
                                 ORIFICE

                                -3 mm —
                               INPUT
Figure 2.5.  CROSS-SECTIONAL VIEW OF SAMPLE INJECTION VALVE.
             scale exaggerated for clarity.
                                                Vertical
on very short columns have not shown any tailing of the injected pulse
due to valve dead volumes.  Several valves have been put through a million
operations with no detectable change in performance.
Figure 2.6.   SEM PHOTOGRAPH OF AN ETCHED SILICON VALVE  SEAT.   (A) Feed-
             through hole to capillary column.   (B) Valve  seating ring.
             (C)  Feedthrough holes to  sample  input  channels.   (D) Valve
             sealing ring.
                                    11

-------
2.3.3  Detector Configuration and Fabrication

After an extensive review of GC detector technology,  a thermal conduc-
tivity detector (TCD) was chosen for use in the integrated gas chromato-
graph.  The advantages of the TCD include fast and general response,  ease
in fabrication and adaptation to the miniature system, ruggedness, and
simplicity of support electronics.  The detector was  designed to be re-
movable from the substrate GC wafer so that it could  be cleaned or re-
placed in the event of detector contamination or failure.

The detector structure is shown in cross section in Figure 2.7.  The gas
flowing through the capillary column exists through a feedthrough hole to
the detector gas channel.  The separate detector chips are mechanically
clamped to the backside of the GC wafer, forming one  side of the detector
channel.  The sensing element is a thin-film nickel resistor which is
supported by a thin Pyrex glass membrane.  This glass membrane supports
the resistor and forms one wall of the gas channel.  The resistor is
heated by an external current source, and its temperature is monitored
by measuring the voltage dropped across the resistor.  When a sample peak
passes by the detector, the thermal conductivity of the carrier gas is
decreased, which causes a corresponding increase in the temperature of
the detector.  This temperature increase causes the detector voltage to
increase which is a measure of the concentration of the sample gas.  The
operation of the detector is covered in more detail in Section 4.3.
                             PYREX  GLASS
                             Si GC SUBSTRATE

                           -ETCHED GAS CHANNEL
                                                            1.5 PYREX
                          1000 X NICKEL FILM                  GLASS
                                                            DETECTOR Si
                       ETCHED  DETECTOR CAVITY  \          / SUBSTRATE
 Figure 2.7.   CROSS-SECTIONAL VIEW OF  DETECTOR  STRUCTURE.  Vertical  scale
              exaggerated  for clarity.


 The detector chips  are separately batch fabricated  using  integrated circuit
 processing techniques  and silicon etches.   The starting wafers  are  200-ym


                                     12

-------
                                  o
thick, (100) silicon.  About 1000 A of Si02 is thermally grown on the wafer
followed by the sputter deposition of about 1.5-pm of Pyrex.  Large holes,
about 300-ym by 700-ym are etched from the back of the wafers to remove
the thermally conductive silicon from the detector region.   About 1000 A
of nickel is evaporated on the Pyrex membrane, defined photolithographi-
cally, and etched to produce the detector resistor.   The chips are then
sawn apart and lead wires are bonded to the resistor to complete the
fabrication process.
                                   13

-------
                          3.  INSTRUMENT SYSTEM
The gas chromatographic air analyzer is a system built around two key
components, the miniature gas chromatograph and a powerful microcomputer.
The chromatograph separates and detects the components of a gaseous sample
and the computer analyzes and displays the output of the chromatograph
and controls the operation of the entire system.

A block diagram of the system is shown in Figure 3.1.  The chromatographic
section includes the integrated GC described in Chapter 2 plus the sample
pump and injection system, a IS-cm^ high pressure helium supply and a
pressure regulator.  The electronics section includes analog electronics
to amplify the detector output, other pressure and temperature sensors,
and digital electronics to process those signals, display the results,
and sequence and control the GC system.
               ATMOSPHERIC
                 SAMPLE
      Figure 3.1.  BLOCK DIAGRAM OF PROTOTYPE GAS ANALYSIS SYSTEM.
3.1  CARRIER GAS SUPPLY

The carrier gas supply is shown in the block digram of Figure 3.2.  The
high pressure helium is passed through a resistive gas path to a high
pressure valve.  The pressure in a surge tank downstream from the valve
is monitored by a pressure transducer.  As carrier gas flows from the
surge tank through the column, the pressure in the surge tank falls very
slowly.  Before the injection of a sample pulse in the GC, the high pres-
sure valve is activated until the pressure in the tank reaches a preset
valve.  The computer then closes the high pressure valve and begins the
sample injection process.  This pressure regulator system works over a

                                    15

-------
             HIGH
          RESISTANCE
             -vw-
                     ON/OFF
                      VALVE

PRESSURE
SENSOR

'
F

COLUMN
RESISTANCE
EXHAUST
         HIGH PRESSURE
         GAS CYLINDER
SURGE
TANK
        Figure 3.2.   SCHEMATIC OF  THE  CARRIER GAS  SUPPLY  SYSTEM.

a wide range of helium source pressures and provides  a nearly constant
carrier gas supply to the GC.  With a  15-cm^ surge tank and  the  nominal
3 Ul/s carrier gas flow,  the carrier supply pressure  will drop about
7 x 1C)2 Pa (0.1 psi) during the course of an analysis.


3.2  SAMPLE INJECTION SYSTEM

The sample injection system chosen for the GC is the  simple  one-valve
system shown in Figure 3.3.  An atmospheric sample is drawn  through a
check valve and through the input  side of the injection valve by a servo-
driven piston pump.   After about 0.6 cm-^ of sample has been  gathered, the
pump reverses, pressurizing the gas sample.  When  the pressure of the
sample reaches a predetermined value,  the computer opens  the injection
valve for a few milliseconds, causing  a few nanoliters of sample to be
injected as a brief plug of gas at the head of the separating column.
The remainder of the sample is bled off from the pump before the next
sample is taken.  The pump is capable  of reaching  pressures  as high as
3.8 x 105 Pa (55 psig) although typical injection  pressures  are  around
1.7 x 105 Pa (25 psig).

3.3  ANALOG ELECTRONICS

The primary purpose of the analog electronics section is  to  condition  the
outputs of the various sensors in the  system, including the  GC detector,
for use by the microprocessor.  The main portion  of the analog electronics
is involved in driving and amplifying  the detector.  A block diagram of
the detector electronics is shown in Figure 3.4.   A constant current
source is used to heat the nickel film resistor.   Another current source
drives a reference resistor to reduce  the effect  of power supply variations
on the output.  The difference between the reference voltage output and
the detector output is amplified by a gain of 5 and then  fed into a high
gain, auto-zeroing amplifier.  The auto-zeroing feature is important  for
the high gain amplifier since the ambient temperature of  the GC  column
is not controlled.  At a total gain of about 5000, the amplified detector
output would drift beyond the dynamic range of the amplifier due to small
changes in ambient temperature.  The high gain amplifier  is  zeroed before
each chromatogram to keep its output within the range of  the analog to
                                    16

-------
       CARRIER
       GAS    *
       INPUT
                              SAMPLE IN
                                     CHECK
                                     VALVE

SERVO
MOTOR








PISTON
PUMP

r»M t




ff\CC

PRESSURE
SENSOR
fiA A,






                       INJECTION
                       VALVE
  -'WV-
RESISTIVE
GAS PATH
»>VENT
                 Figure 3.3.   SAMPLE INJECTION SYSTEM.
                          DETECTOR ELECTRONICS
        DETECTOR
        CURRENT
        SOURCES
                                x1  AMP
                                           LOW GAIN
                                           OUTPUT TO A/D
                                                        HIGH GAIN
                                                        OUTPUT TO A/D
                                 AUTO-ZERO
                                 SWITCH
                                         OFFSET
                                         INTEGRATOR
Figure 3.4.  SCHEMATIC OF DETECTOR CURRENT SOURCE AND DETECTOR AMPLIFIERS.

                                  17

-------
digital (A/D) converter.   A low gain amplifier  is  also  connected  directly
to the detector to monitor the 1-volt air peak  signals, which  saturate
the high gain amplifier.

The A/D converter is the  primary way that the computer  obtains information
from the analog electronics.   It is switched among the  various pressure
sensor amplifier outputs  and other signals as shown in  Figure  3.5.   The
pressure sensor outputs are used to control the operation of the  carrier
gas supply and the sample injection system.  A  thermistor is used to
measure the column temperature in order to calculate the  expected reten-
tion time of the sample peaks.
            CARRIER
            GAS
            PRESSURE
            SENSOR
            SAMPLE
            INJECTION
            PRESSURE
            SENSOR
                                           CONTROL
           Figure 3.5.  A/D CONVERTER AND MULTIPLEXER SYSTEM.
                                    18

-------
The remainder of the analog electronics allows the computer to drive the
sample pump, injection valve and pressure regulator valve.  In addition
there are voltage references and regulators to stabilize the battery
voltages and provide drive signals for the pressure and temperature
sensors.

3.4  DIGITAL ELECTRONICS

The digital electronics package consists of a microprocessor chip, its
associated program and data memories, a keyboard interface, and a liquid
crystal display, as shown in Figure 3.6.
                             COMPUTER SYSTEM

POWER
DOWN
.I
TIME

CLOCK

ALARM





Z-80
M
1
C
R
0
P
R
0
C
E
S
S
0
R
DATA
BUS
{> OADDRESS{
1 || BUS
1TTT
— -s


I)
\

A— D
CON-
VERTER


A
D
KEY-
BOARD
INTER-
FACE


111

„'
DIS-
PLAY


CONTROL LINES
             Figure 3.6.  DIGITAL ELECTRONICS BLOCK DIAGRAM.
The microprocessor chosen for the system is a Zilog Z-80a.  It is a
relatively high speed, 8-bit device with an advanced instruction set.
Since the power dissipation of this device is relatively high (1 Watt),
a unique powerdown system is used to minimize the time during which power
is applied to the high power consuming portions of the system.  It is
anticipated that functionally equivalent Z-80 devices will be available
in 1981 which will dissipate about 20% of the present power.  This advance
will be a significant help in extending the battery life of the system or
reducing the present battery size.

The computer's memory is divided into two sections, permanent program
memory and volatile random access memory.  The program for the computer
is held in 10 kilo-bytes of electrically programmable read-only memory,
EPROM.  Since there is no reason to change the program during its use,
the EPROM is programmed once and will store the program almost indefinitely.
The volatile memory is used by the computer as a scratch pad for calcula-
tions and to store system parameters, gas parameters, and data on the

                                    19

-------
concentrations which it has determined.  These results are held in 2
kilo-bytes of extremely low power CMOS memory which is powered continu-
ously, even when the machine is "turned off".  The keyboard allows the
user to completely control the operation of the instrument by displaying
and by modifying the parameters which are used by the computer to control
the instrument.

In addition to the interfaces with the analog electronics, the digital
system includes a 16 function keyboard and a 4-digit liquid crystal dis-
play.  The keyboard offers multiple levels of control depending upon key
sequence.  It can be used to retrieve the sample gas concentrations, in-
put calibration data to the computer, control the operation of the machine,
and retrieve other important parameters such as carrier gas pressure,
temperature, and peak retention times.  The display is used to output all
computer data and to check the accuracy of inputted calibration parameters.

3.5  PACKAGE

The complete instrument is housed in a package about 20 cm x 22 cm x 10 cm
(8" x 9" x 4") which weighs 2.8 kg (6.2 Ibs).  The details of the physical
layout of the package are shown in Figure 3.7 and in the photograph,
Figure 3.8.  A majority of the volume and weight of the present package is
taken up by batteries and electronics boards.  It is anticipated that the
present battery weight of 765 gm, which is 27% of the total, will be re-
duced by a factor of about 2 when the lower power digital components
become available.  The volume associated with the electronics can be re-
duced by at least a factor of 2 by using multilayer printed circuit boards.
Since the digital system design is still subject to modification as im-
proved components become available, wire-wrap techniques, which are more
easily changed than printed circuit boards, are presently used in the
prototype instruments.  It is further anticipated that new interface and
memory components will become available which could be used to further
reduce the size of the electronics boards.

The computer controls the entire system, so there is no need for user
adjustable flow controls, zero controls, or meters to read.  The front
panel only contains a keyboard to interface with the computer, a mode
switch to select automatic or survey mode, and the digital display.  The
only maintenance tasks are replacing the high pressure helium cartridges
and recharging the batteries.  The high pressure helium cartridges are
changed by removing the empty cartridge from a holder on the bottom of the
unit and placing a full cartridge in the holder.  A screw is then rotated,
forcing the cartridge against a puncturing needle and a sealing gasket.
The entire operation takes less than 1 minute, and the helium cartridges
are available from a commercial source for about $1.75 in quantities of
1000.  Since the cartridge lasts for at least 4 days of operation, the
helium operating costs are quite modest.

The complete GC subsystem occupies about 15% of the present instrument
package and a similar percentage of the weight.  The GC system could be
separated from the electronics, if a worker-wearable version of the system
is desired.  The GC could then be worn by the worker near the breathing zone,
                                    20

-------
         ANALOG CIRCUITRY
         TOP VIEW: MIDDLE LAYER
                                                          DIGITAL ELECTRONICS
                                                          AND COMPUTER BOARD
                                                                      MINIATURE
                                                                         GC
                                                                       HOUSING
                                                         TOP VIEW: TOP LAYER
                  \ HIGH
                  v PRESSURE
                   VALVE
        TOP VIEW: BOTTOM LAYER
                                                               -20cm-
DIGITAL BOARD LEVEL


ANALOG BOARD LEVEL

UPSURGE TANK ")


V
\\
f\\ F3
, .y'ffipJsIy ,,





(4

COLUMN
HOUSING
CHECK
VALVE
SYRINGE
PUMP
            GC'
          WAFER
                SOLENOID

INTERNAL VIEW FROM FRONT
                         SUPPORT
Figure  3.7.   LAYOUT  OF  THE PROTOTYPE MINIATURE GAS ANALYSIS  SYSTEM.
                                         21

-------
Figure 3.8.  PHOTOGRAPH OF THE INTERIOR OF THE PROTOTYPE INSTRUMENT.  The
             GC is in the lower right and the digital computer board oc-
             cupies most of the remaining top layer.
and the electronics, batteries, and carrier gas supply could be carried
on the belt.  After the battery and electronics systems are reduced in
size and weight, the total system will be much easier to wear for an 8-
hour shift.
                                   22

-------
                             4.  PERFORMANCE
The performance of a complicated instrument such as the miniature gas
chromatograph is dependent on both the individual performance of the
separate elements and the manner in which they are interconnected to form
the system.  The following section will concentrate on the system perfor-
mance, with the succeeding sections devoted to the performance of the
individual chromatographic, detector and computer sections.

4.1  OVERALL INSTRUMENT PERFORMANCE

The basic measure of performance used for this instrument is the overall
system reproducibility, defined here as the ability of the instrument to
measure and display the concentration of a standard gaseous sample.  Most
of the reproducibility tests have used dilutions from commercially avail-
able calibrated gas samples of pentane and hexane at original concentra-
tions of 1000 ppm.  A large number of analyses are performed on a sample
gas, and the mean and standard deviation of the distribution of results
are calculated.  Since the analyses are performed in less than one minute,
a rather large number of essentially independent measurements can be made.
The results of a typical test consisting of 50 analyses showed the instru-
ment's measurement of a sample containing both pentane and hexane to have
a percent standard deviation of 2.5% for pentane and for hexane, to give
a reproducibility at the 95% confidence level of plus or minus 5% at
1000 ppm concentrations.  In such a test at approximately room temperature,
the pentane retention time is about 3.3 seconds, closely following the
air peak, while the hexane at 4.6 seconds is roughly in the middle range
of retention times (4 to 8 s) for this instrument.

A series of tests of reproducibility were performed on the prototype instru-
ment.  Each test involved taking 50 samples of a gas mixture, and the mean
and standard deviation of the concentration results were determined.  The
gas samples were prepared by successive dilutions of a 1000 ppm standard
gas, and due to uncertainties in the dilution process, the dilute gas
samples are of unknown absolute accuracy.  From the more than 3000 analyses
which were performed on the prototype instrument a representative set of
50 consecutive analyses of diluted samples of pentane and hexane at about
50 ppm has been selected and the individual results are shown in Table 4.1.
This table lists the computer calculated concentrations of pentane and
hexane in ppm and the measured air peak heights in mV from an external
oscilloscope.   A typical chromatogram from this series is also shown in
Figure 4.1.  The calculated percent standard deviation for pentane is
3.85% with a mean of 49.7 ppm, and for hexane the mean is 67.9 ppm with a
4.89% standard deviation.  The maximum and minimum calculated concentra-
tions are within +2 standard deviations of the mean.

                                    23

-------
             Table  4.1.  Results of reproducibility test.
Run
#
1



5




10




15




20




25
Pentane
Hexane
Pentane
(ppm)
51.6
51.6
46.3
48.6
51.3
48.3
53.1
50.7
49.6
48.0
51.8
46.1
50.5
51.6
52.1
47.7
50.6
50.1
46.5
50.2
47.1
47.5
46.2
48.1
51.5
mean:
mean:
Hexane
(ppm)
66.1
70.1
65.9
66.7
61.7
66.3
68.3
67.8
70.0
66.8
72.7
72.0
68.9
71.9
71.4
65.7
67.2
70.7
67.1
67.8
71.6
62.7
69.0
71.5
73.9
49.75 ppm
67.86 ppm
Air
(mV)
1962
1924
1902
1926
1926
1932
1950
1918
1918
1912
1930
1922
1896
1934
1930
1914
1898
1918
1916
1930
1926
1944
1924
1932
1910


Retention times:






Air 2.7 s
Pentane 3.36
Hexane 4.55

s
s
Run
#
26



30




35




40




45




50
Pentane Hexane
(ppm)
50.7
50.0
51.7
48.5
49.5
51.8
49.8
48.1
51.2
51.1
48.9
49.4
53.0
49.8
51.6
51.2
50.9
50.6
46.3
47.2
50.2
52.7
48.8
48.4
49.7
(ppm)
67.0
67.4
69.5
63.5
68.7
66.3
66.9
69.1
73.4
63.1
65.4
65.9
74.9
67.0
67.6
70.1
64.3
64.3
66.8
66.1
62.6
68.5
60.3
65.6
65.0
Pentane standard deviation:
Hexane
Column
OV-101
Helium

standard deviation:
temperature
stationary
carrier gas

27°C
phase


Air
(mV)
1900
1924
1902
1918
1942
1908
1930
1924
1910
1904
1912
1910
1916
1916
1902
1902
1908
1892
1914
1914
1908
1888
1908
1894
1898
3.85%
4.89%




This use of the percent standard deviation is a valid measure of the re-
producibility of the instrument for single measurements, that is, any
single reading can be expected to be within +2 standard deviations of the
mean with a 95% confidence level.  When the instrument is used to repeat-
edly sample the atmosphere or a gas sample, the overall reproducibility of
the final time weighted average (TWA) figure will have substantially
smaller percentage error, assuming all other conditions are constant.  The
percentage error will theoretically decrease by a factor of /n, where n
is the total number of samples.  During an 8-hour day, the instrument can
                                    24

-------
Figure 4.1.  TYPICAL CHROMATOGRAM FROM THE REPRODUCIBILITY TESTS.  Peaks
             are  (A) air,  (B) pentane, (C) dichloromethane,  (D) hexane,
             (E) benzene,  and (F) trichloroethane.  Sample compounds are
             approximately 50 ppm concentrations.


take more than 400 samples, so that the final TWA figures can be expected
to be considerably better  than the single reading percent standard devia-
tions.  For example, for the pentane results shown in Table  4.1, the mean
for each column of 25 analyses are found to be 49.468 ppm and 50.044 ppm.
These two results have a standard deviation of only 0.82%, which is quite
nearly a factor of /25~ better than the single measurement error of 3.84%
standard deviation.

The results of a series of tests with pentane, hexane, and carbon tetra-
chloride with concentrations from 10 to 1000 ppm are shown in Figure 4.2.
The reproducibility of pentane is better overall than hexane, for example,
because the retention time of pentane is shorter than that of hexane,
which means that for equal concentrations, the peak height of pentane will
be significantly higher than that of hexane.  The increased peak height
improves the signal to noise ratio, which improves reproducibility, es-
pecially at small concentrations.  In these tests at room temperature, the
respective retention times for air, pentane, hexane, and CC14 were about
2.7, 3.3, 4.6,  and 6.0 seconds.   Since the pentane peak is only about 500
ms from the air peak, pentane peaks less than 20 ppm are lost in the tail
of the air peak by the present computer program.

The reproducibility of the instrument at larger concentrations of about
2.5% standard deviation is quite acceptable.  It is important to identify
the sources of  error at the different concentration levels in order to
determine possible future improvements in system performance.  The sources

                                    25

-------
 100
                       10                   100
                NOMINAL GAS CONCENTRATION (ppm)
1000
           Figure 4.2.   SYSTEM REPRODUCIBILITY VS SAMPLE SIZE.
of error fall roughly into  four  categories, GC system errors, detector
noise and nonlinearities, quantization errors in the A/D converter, and
errors in the computer algorithms used to compute the peak areas.  Each
of these sources of error has  received some study, and the performance in
each area can undoubtedly be improved with additional work.

The GC process itself probably contributes the least amount of uncertainty
to the system reproducibility.   If  identical samples of constant gas con-
centration are injected at  the head of the column, then identical peaks
are expected at the output  if  identical  temperature and flow conditions
are maintained.  Possible sources of error are in the sample collection
and injection system, however.   Some of  these errors, such as slightly
different sample injection  volumes  are taken into account in the final
concentration calculation.   There are some dead volumes in the sample in-
let region which result in  a finite rise time of about 4 sampling intervals
to reach 95% of a step change  in concentration.  In the reproducibility
test it is assumed that the GC system is basically not responsible for any
of the system error.

The GC detector, as discussed  in Section 4.3.2, is slightly nonlinear and
has some low level noise.   In  this  prototype the detector noise and the
analog electronics noise is about 2-yV peak-to-peak, referred to the

                                    26

-------
detector, at nominal operating conditions over a bandwidth of 0.1 to 30
Hz.  This detector noise is thought to be due to extremely small flow in-
stabilities in the detector cavity, which causes minute changes in
detector operating temperature.  This detector noise only becomes impor-
tant at the lower peak concentrations.  The unit is generally operated so
that the output peak height, referred to the detector, is about 0.5 to
1.0 yV/ppm for peaks such as hexane and chloroform.  In the reproduci-
bility tests, the injection pressure was set so that sufficient sample
was injected to obtain about a 2000-mV air peak and a sensitivity of about
0.7 UV/ppm for hexane.

The predominant source of error in the system at lower concentrations is
caused by the A/D converter.  The present 10-bit A/D converter has a
minimum resolution of 5 mV.  Under the conditions of the reproducibility
test, a 20 ppm hexane peak is 70 mV, which is about 14 A/D levels in
amplitude.  This lack of amplitude resolution introduces significant error
in the peak area calculation.

When measuring small gas concentrations, it is possible to improve repro-
ducibility by a factor of about 2 in the sub-10 to 50 ppm range by in-
creasing the size of the peaks by injecting a sample at higher injection
pressure.  This increases the amplitude of the peaks, which improves the
signal to noise (S/N) ratio.  The only disadvantage of this procedure is
that a 1000 ppm sample of, for example, pentane would be so large as to
be out of the dynamic range of the A/D converter.  Thus, at higher injec-
tion pressures it would not be possible to simultaneously measure a 1000
ppm hexane peak and a 10 ppm CCl^ peak.

Peaks in the 20 ppm range are still well above the detector and electronics
noise level so the reproducibility at these levels could be improved by
increasing the gain of the amplifier or by increasing the resolution of
the A/D converter.  The amplifier gain has been selected such that a 1000
ppm pentane peak is just in the maximum range of the converter.  It would
be possible to change the gain of the amplifier during the chromatograph,
but this would introduce additional complications in the computer program,
which would probably unacceptably slow down the program, and would further
reduce the sampling rate.  A 12-bit A/D converter is presently being tested
and is most likely the quickest way to achieve better performance, and when
faster, low power computers become available, the program can be modified
to handle amplifier gain changes.

It appears that the main source of inaccuracy in the computer system is
the finite sampling rate.  The computer has only enough time to take about
30 samples during the time equivalent to two standard deviations of the
roughly gaussian output peaks.  This causes imprecision in the determination
of the peak standard deviation which is used to calculate the peak area,
which is further used to determine the peak concentration.  There are two
ways to reduce this error.  One is to sample faster, which would require
faster and more sophisticated computers, and the other is to use an addi-
tional interpolating filter to achieve higher precision than the sampling
frequency allows.   Both approaches may be used in more advanced versions
of the instrument.
                                    27

-------
4.1.1  Analog Electronics Performance

The computer system, since it is digital, is immune to temperature and
battery voltage changes, as long as they remain within the operating
range of the devices.  The analog electronics, however, is subject to
temperature and power supply variations.  Most of these variations can
be minimized by appropriate circuit design, and the use of resistors and
capacitors with inherently low temperature coefficients.  The primary
reference element in the analog electronics is a voltage source which is
stable to within 0.08% over the operating temperature range of 0 to 45°C.

The reduced results of the analog electronics test are shown in Table 4.2.
This test measured the changes over a temperature range of from 25 to 50°C
and over the maximum voltage range of the two batteries of +5 to +7 V and
-14 to -10 V.  The two individual effects are listed along with the total
extreme over both temperature and voltage.  The portions of the circuit
measured are the most important ones for the analytical performance of the
     Circuit
Table 4.2.  Analog electronics performance.

      Drift over Temp    Drift over Supply
       25°C to 50°C       -14,7 to -10,5
Total Extreme
   T and V
Thermistor amp
equiv. drift
Detector amp
zero drift
Pressure output
zero drift
Pressure output
30 psi
Swept time number
10 seconds
Detector current
0.064°C
2.2 mV
0.58 psi
0.17%
0.63%
1.08%
0.034°C
0.34 mV
0.07 psi
0.37%
0.72%
0.65%
0.10°C
4.6 mV
0.70 psi
0.37%
1.6%
1.65%
instrument.  The thermistor amplifier error is the output voltage error due
to the reference and amplifier converted to an equivalent temperature
measurement error.  The detector amplifier zero drift is the total output
voltage change of the xl amplifier with a metal film resistor in place of
the detector.  This 4.6-mV total change is less than 1 A/D level and in-
cludes the effects of current source changes on the input bridge, the
temperature change of the detector reference resistor, and the xl amplifier
drift.  Even though this change is small, after an additional gain of 5000
the drift would be considerable.  The high gain amplifier is zeroed before
each chromatogram to reduce the effect of these drifts on the high gain
output.  If the instrument is subjected to a 20°C step change in tempera-
ture, the resulting high gain output drift is about 1 V during a chromato-
gram.  The pressure errors include the temperature effects of the pressure

                                    28

-------
sensors along with the bridge driving voltage and the transducer ampli-
fiers.  The swept time error is the change in the number of clock pulses
generated over a 10 second period by the swept time circuits.  The  1.6%
total error corresponds to less than a standard deviation of one of the
output peaks.  The relatively large (1.65%) total error in detector cur-
rent is of little importance, since.the detector sensitivity is changed
little by this difference, and the concentration calculation performed by
the computer is concerned only with the relative sizes of the air and
sample peaks, so sensitivity changes cancel when the ratio is taken.

4.2  GC COLUMN PERFORMANCE

The performance of any GC column is determined by a myriad of parameters,
including the thermodynamic properties of gases and liquids, and the
geometry of the column and its effects on gas flow.  In general, the most
important factors are the stationary phase used and the geometry of the
GC column.  Since the start of gas chromatography in the 1950's, there
have been numerous attempts to provide a complete theoretical description
of the gas chromatography process.  The geometry of the miniature chroma-
tograph is sufficiently similar to that of conventional instruments that
the GC theory can be used to provide clues to help optimize the analytical
performance of the instrument.  Unfortunately, the mathematical descrip-
tions of the actions of the GC linings do not indicate how to actually
line a column to produce maximally efficient linings.  The lining tech-
niques must be separately developed for the modified geometry of the mini-
ature GC, and the theory can then be used to optimize the size of the
columns to perform specific separations.

4.2.1  Column Linings

The stationary phases used thus far in the miniature capillary columns
have been the same substances as those used in conventional gas chromato-
graphy.  Modified lining techniques have been developed which achieve
reproducible, uniform distributions of the stationary phases on the column
walls.  Most of the separations performed with the miniature capillary
columns were achieved with a silicone oil, OV-101, as the stationary phase.
The following sections describe the column treatment and lining procedure
for the OV-101 liquid phase.

Column Pretreatment—
The construction of the integrated capillary column is such that the cap-
illary bottom and side walls are silicon while the top plate is Pyrex
glass; however, the silicon surface is coated with a layer of silicon
dioxide (Si02) several tens of Angstroms thick which is formed during the
anodic bonding process.  The Si02 and glass surfaces are naturally covered
with a high density of silyl ether and hydroxyl groups.  Treatment of
these surfaces with an aqueous solution of a strong acid (e.g. concentrated
HC1)  converts many of the surface  ethers into hydroxyl groups.
                                               Si-OH

                                               Si-OH
                                    29

-------
These surface hydroxyl groups will react with organosilane compounds of
the form RS1X3 or R3SiX (where R is an alkyl group and X is either a
chloride or an alkoxide group) to form a Si-O-Si linkage between the sili-
con and the silane as


                                           R
                     Si-OH + R SiX -* Si-0-Si-R + HX
                                           R


We have employed this chemistry to react trimethylchlorosilane (^3)3 SiCl
with the column surface.  The reaction is carried out by forcing the un-
diluted silane through the column for about 10 min at 70°C.  The capillary
is then washed with water and methanol and blown dry.  This procedure
serves to convert the OH groups which populate the surface of the column
and to provide a monomolecular layer to which the stationary phase can
adhere.

Dynamic Column Lining—
The liquid stationary phase can be applied to the miniature capillary walls
by means of a conventional dynamic coating procedure [3].  The OV-101
liquid is dissolved in a very volatile solvent such as chloroform, and a
plug of the mixture is forced through the column.  If the speed of the
moving plug is kept constant, a uniformly thick layer of OV-101 is left on
the column walls.  The approximate thickness (t) of the layer is given by
                                uoo/U
where C is the concentration of the solution (%v/v), r is the radius of
curvature of the capillary tube, y is the average velocity of the plug of
lining solution, r| is the solution's viscosity, and y is i-ts surface ten-
sion.

Although this lining procedure is common practice with conventional capil-
lary GC columns, it is less than ideal for the miniature capillaries since
they have extremely smooth surfaces and nonconstant radii of curvature.
The resulting distribution of stationary phase appears to be nonuniform
both across the width of the capillary channel and along its length.  The
lining distribution is unstable with time and the OV-101 bleeds out of the
column.

The distribution and performance of the column lining can be significantly
improved with the addition of a small percentage of 7-10 A Si02 particles
into the OV-101 liquid phase.  These particles tend to increase the
stability of the thin liquid layer and result in greater surface area of
liquid phase per unit length of column, an attribute which increases the
column's ability to separate gases.
All of the miniature capillary columns which are fabricated now include
                                   nai

                                    30
SiO_ particles in the liquid stationary phase.  A column lined with OV-101

-------
and Si02 particles was operated with constantly flowing carrier gas for
more than 6 months.  No attempt was made to condition the column, and
numerous temperature cycles were performed, but the relative retention
times of the peaks changed less than 10% over the length of the test, and
no detectable amount of column lining was blown into the detector channel.

Static Column Lining—
With the miniature capillary column as well as with conventional columns,
the static coating procedure [4] is replacing the dynamic method because
it yields more reproducible and uniform stationary phase distributions.
The OV-101 liquid is dissolved in chloroform along with S102 particles and
the capillary is totally filled with the mixture.  The output end of the
capillary is then sealed off and the miniature GC column is put in a vacuum
chamber at less than 1 Torr pressure.  The chloroform in the capillary
evaporates slowly from the head of the column leaving the OV-101/Si02
residue coating the capillary walls.  The evaporation rate is slow and
constant and all of the OV-101 remains in the column.  This procedure
gives very uniform lining distributions in which the exact weight of
liquid phase per unit length of column is known.

4.2.2  GC Separations

One of the important features of the miniature gas chromatograph is its
very,rapid separation of the gases of interest.  As a simple example of
its capabilities, an oscilloscope photograph of the amplified detector
output, or chromatogram, is shown in Figure 4.3.  The sample gas consisted
of a mixture of nitrogen, pentane, and hexane vapors and was injected at
time zero.  The nitrogen has no affinity for the OV-101 column lining so
Fifugre 4.3.  OSCILLOSCOPE PHOTOGRAPH  OF THE AMPLIFIED DETECTOR  OUTPUT  OR
              CHROMATOGRAM.  The sample was injected at the beginning of
              the sweep.  The peaks are  (1) nitrogen,  (2) n-pentane, and
              (3) n-hexane.
                                    31

-------
It emerged from the column after 2.6 seconds, a time determined solely by
the helium carrier gas velocity.  The finite base width of the nitrogen
peak ("air peak") is caused by gaseous diffusion of the nitrogen in the
carrier gas, since the injected peak was only 5.0-ms wide and there are
no appreciable mixing or detector time constants.  The next peak at about
3.3 seconds is the pentane, followed at 4.6 seconds by the hexane.  The
increased retention times are due to the times spent by the pentane and
hexane adsorbed on the column lining, and the increased base widths of
these peaks are due both to gaseous diffusion and diffusion of the species
in the column lining.  It is preferable to keep the peak spreading to a
minimum, since this will facilitate the separation of more components of
a mixture and will maximize the peak height which increases the signal to
noise ratio of the detector.  The complete separation was performed in
less than 5 seconds, which is significantly less than the several minutes
which a conventional GC system might take to give the same separation.

Figure 4.4 shows another output chromatogram using OV-101 as the stationary
phase.  This example shows the ability of the device to separate very
similar branched and straight chain hydrocarbons  as  well as chlorinated
hydrocarbons.  Again, the relatively short length of the miniature capil-
lary column (1.5 meters) enabled the separation to be performed in less
than 10 seconds.  The analysis time is shorter than a typical convention-
al GC, but the separation of these closely related chemicals which was
achieved by the miniature GC is not as good as could be achieved with a
standard column.  By applying capillary GC theory the separating power of
the miniature GC can be characterized.
Figure  4.4.  MINIATURE GC CHROMATOGRAM USING  OV-101 STATIONARY PHASE.
             Peak  (1) nitrogen,  (2) n-pentane,  (3) 3-methyIpentane,  (4)
             n-hexane and chloroform,  (5) 2,4-dimethyIpentane,   (6)  111-
             trichloroethane,  (7) cyclohexane,  and (8) n-heptane.
                                    32

-------
The sketch of an idealized chromatogram shown in Figure 4.5 defines the
parameters which characterize an output peak.  The retention times from
sample injection to the maximum of each peak for the unretained air peak
and the sample peak are tyi and tR respectively.  An important measure of
the operation of the column is the length of time that the sample is
retained by the stationary phase.  This is the adjusted retention time
t^, given by t^ = t^ - tjj.  For a gaussian peak with standard deviation a,
the peak basewidth is defined as W = 4a.
 Figure 4.5.   IDEALIZED  CHROMATOGRAM  SHOWING RETENTION TIMES AND PEAK WIDTHS,
 One practical measure of  column efficiency  is  the effective number  of
 theoretical plates  (N) in the  column,
                              N =  16
VwJ
(i)
 It  is desirable  for N  to be  large,  indicating  a well  retained yet  very
 narrow peak.  The number of  effective plates calculated  for  the  peaks in
 Figure 4.4  ranges from 385 for peak #2  to  2300 for peak  #8.  For 30-m long
 and efficient standard capillary  columns,  the  number  of  effective  plates
 typically ranges between 10,000 and 100,000.   The improvement in perfor-
 mance of the conventional column  is due to its greater length and  the
 superiority of its stationary phase distribution.

                                     33

-------
The utility of the number of effective plates lies in its ability to
predict if a column is efficient enough to resolve two peaks whose ad-
justed retention times tjQ and tj^ are very close.  The standard measure
of resolution of two peaks is defined as
                                  tf  - t1
                             R = .>   'Rl
                                       W2)/2
A resolution value of R = 1.5 results in peaks separated as shown in Figure
4.5B.  The number of effective plates, N, needed for a column to separate
two peaks with a resolution of R = 1.5 can be determined from
                             N = 32
                tR2
where       a = — ; —
                tRl
As an example, consider peak #4 of Figure 4.4.  This peak is really two
unresolved peaks of chloroform and hexane for which a = 1.05.  Equation (3)
predicts that the column would need N = 14,122 plates to just separate
those two vapors.  Since N was only 1067 plates for this particular column,
those two vapors were not separated and thus appeared as only a single
peak.

The ability of the miniature capillary column to separate peaks is deter-
mined by the stationary phase, the column geometry, and the column opera-
ting conditions.  The current columns 1.5-m long by 40-ym by 175-ym wide
and lined with OV-101, are only about 20% efficient.  That is, the columns
exhibit only 20% of the maximum number of theoretical plates possible for
a column with those dimensions.  Thus, improvements in the stationary phase
composition and lining techniques should result in a significant improve-
ment in column performance.  Once the stationary phases have been improved,
the dimensions of the capillary columns can be optimized for given appli-
cations and specific separations.

4.2.3  Optimization of the GC Column

The number of effective theoretical plates that a column possesses for any
gas is related to another parameter, the plate height (H) by
                             H ~
where L is the column length and k is the partition ratio for a gas of
interest and is given by

                                    34

-------
The plate height of an open-tubular capillary column with rectangular
cross-section has been related to the column parameters by Golay [5J as
                           H = - + (C, + C )u                         (4)
                               u     a    g

where
     B = 2D
           g
              3 2
         3d + k)   D
                9k + 51k2
             105(1 + k) D
                         o
                                                                      (6)
and
    Z_ = 1/2 of the etched column depth

     u = carrier gas velocity

    D  = diffusion coefficient of solute vapor in carrier gas
     O

    D0 = diffusion coefficient of solute vapor in liquid stationary
     JG      -
           phase
The partition ratio k used in the above equations is a function of the
column height and width.  It is related to a more fundamental parameter,
the partition coefficient K, by the equation
*
 The partition coefficient is defined as the concentration of solute vapor
 in the stationary phase divided by the concentration in the mobile phase.
 It is only a function of the thermodynamic properties of the solute vapor
 and the stationary phase.  Its exponential dependence on temperature makes
 it the most temperature-dependent parameter, and thus facilitates the con-
 trol of column performance with simple temperature adjustments.

                                    35

-------
                                d(2Z  + Y)
                            k	f-^— K                          (7)
where

     d = thickness of the layer of liquid phase lining the column

     Y = column width
The B/u term in Equation (4) accounts for peak spreading due to longitu-
dinal diffusion of vapor molecules (diffusion parallel to the column axis).
longitudinal diffusion occurs only while the vapor molecules are in the
mobile phase and can be characterized by a diffusion coefficient Dg for
the specific vapor in a specific carrier gas.

The Cg and G£ terms can be regarded as the "resistance to mass transfer"
of the solute vapor in the mobile and stationary phases, respectively.
They represent the departure from the ideal condition of instantaneous
equilibrium between the vapor in the gas and liquid phases.  Cg is asso-
ciated specifically with the parabolic carrier gas profile across the
column's vertical dimensions.  The faster moving vapor molecules in the
center of the gas stream take longer to diffuse to the stationary phase
than do their slower counterparts near the column walls.  Cg can be re-
duced by reducing the column depth (thus shortening the diffusion path
for any vapor molecule) or by increasing Dg (speeding up the rate of dif-
fusion) .

The C^ term accounts for the diffusion of vapor molecules into the liquid
stationary phase.  Once a vapor molecule impinges upon the liquid surface,
it tends to migrate into the liquid phase as the vapor concentration in-
creases locally in the mobile phase.   As the vapor peak passes by and the
mobile phase concentration decreases, the process is reversed, and the
vapor molecules in the liquid diffuse back to the surface and transfer to
the gas stream.  The G£ term can be reduced by increasing the vapor dif-
fusion rate through the liquid phase (increasing D&) or by reducing the
liquid layer thickness d.

Since the Equation (4) for plate height contains the carrier gas velocity
in both the numerator and denominator, it can be minimized with respect to
u.  Such an optimization yields
                              opt    / C  + Cn
                               v    -V  g    1
                          H    =2 /B(C  + Cj
                           opt         g    1
                                    36

-------
A Golay plot of H as a function of carrier velocity can be generated
experimentally, and a typical plot for a miniature column is shown in
Figure 4.6.  The values of B, Cg, and Cjj, as well as the optimum carrier
velocity can be derived from this plot.  Unfortunately, almost all of
the terms of Equation (4) are functions of column geometry and partition
ratio parameters; however by assuming typical column operating conditions
and plotting Equation (4) as a function of a specific column parameter,
some insight can be gained into that parameter's effect on the operation
of the column.
          £
          I
          UJ
          I
          UJ
             0.5
             0.4
             0.3
             0.2
             0.1
                        0.2       0.4      0.6       0.8
                        CARRIER GAS VELOCITY (msec)
1.0
Figure 4.6.  GOLAY PLOT OF MINIATURE CHROMATOGRAPHIC COLUMN WITH OV-101
             COLUMN LINING.
The most signficant geometrical feature of the miniature capillary column
is its roughly rectangular cross section with very shallow depth  (Z).  The
calculated values of HQpt and NOpt are plotted in Figure 4.7 as a function
of column depth.  The carrier gas velocity required to achieve these opti-
mum values is plotted in Figure 4.8 as a function of column depth.   It is
obvious from Figure 4.7 that the best column performance (largest number
of effective plates, N) is realized as the etched column depth is decreased.
However, the price to be paid for increasing the column efficiency by re-
ducing its etched depth is the much higher pressure drop across the  column
which is required to achieve the optimum carrier gas velocity.  Given a
maximum operating pressure, Equation (4) can once again be used to show
how N varies with column depth.  Such curves are plotted in Figure 4.9 for
several values of maximum carrier pressure.  Clearly there is an optimum
column depth for each value of carrier pressure.
                                    37

-------
            10
X
tr

o

          cc
          2
          UJ
6
                       10
                      20        30
                        z0 (microns)
                                       40
                                                               O.I
                                                               0.08
                                                                   o
                                                               0.06
                                                                   o
                                                               0.04
                                                               0,02
50
Figure 4.7.   OPTIMUM PLATE HEIGHT AND SEPARATION  FACTOR AS FUNCTIONS
              OF COLUMN DEPTH  (ZQ).
            100
            80
            60
          u
          _&40
            20
                        10
                       20        30
                         z0 (microns)
                                        40
                                                              50
   Figure  4.8.   OPTIMUM CARRIER GAS VELOCITY AS  A FUNCTION OF  Zr
                                    38

-------
                                   INPUT PRESSURE P( (psig)
                                   VARIED AS PARAMETER
                               40       60
                                 z  (microns)
100
Figure 4.9.  NUMBER OF EFFECTIVE PLATES AS A FUNCTION OF MAXIMUM ALLOWABLE
             CARRIER GAS PRESSURE AND ETCHED DEPTH.
The preceding example illustrates how one of the many parameters which  con-
trol the operation of the miniature capillary column can be optimized to
increase the efficiency of the column.  When such analyses are performed
for the other column parameters, the following guidelines are suggested for
optimizing the design of the miniature column with a specific stationary
phase for the separation of a known group of sample vapors.

     (1)  The maximum allowable carrier gas pressure should be used.
          As shown in Figure 4.9, the peak value of the number of  ef-
          fective plates increases with input pressure.

     (2)  The column should be optimized for the separation of the
          vapors with the smallest retention times or for the most
          difficult separation.

     (3)  The column length should be as long as possible if the maxi-
          mum separation is desired.
                                    39

-------
     (4)  Once the values of the partition coefficient, column length,
          and carrier pressure are known, a plot of Equation (4) will
          indicate the column depth which maximizes the separating
          power of the column.

Such an optimization of the miniature capillary column must be tailored
to fit many other requirements imposed by the rest of the instrument sys-
tems.  For example, the column dimensions and the carrier gas velocity
determine the rate of consumption of carrier gas and thus influence the
size of the helium supply tank.  Long columns are desirable to perform dif-
ficult separations, yet the resulting analysis time may be prohibitively
long.  Very short columns for rapid separations are too fast for the com-
puter to keep up with in real time, and thin column linings require very
small sample injections which may limit system detectivity.  Many such
trade-offs exist, making system optimizations a difficult task.

The miniature GC columns which are currently in use are 1.5-m in length
with 40-ym x 175-ym cross sections.  There has been no comprehensive ef-
fort to optimize the geometry of these columns since the stationary phase
distribution and performance are not under good control.  The choice of
some of the column dimensions and operating parameters has been dictated
by reliable valve operating pressures, desired anslysis times, and by
computer speed.

The column length of 1.5-m was arbitrarily chosen at approximately half of
the maximum length that we can currently fabricate.  The 40-ym column depth
gives air peak retention times of 1 to 3 seconds over carrier pressures
of 30 to 10 psig.  The Golay plot shown in Figure 4.6 indicates that car-
rier gas velocities between 61 and 87 cm/s yield the minimum plate height
and hence the best column performance for the stationary phase distribution
which is currently achieved.  The velocity of 61 cm/s, which corresponds
to a helium carrier gas pressure of 19 psig, was chosen so that the re-
sulting air peaks and little-retained gas peaks would have base widths of
at least 130 ms.  Such peaks are of the minimum width which can be ade-
quately sampled by the microcomputer.

The nonpolar columns which are currently used are only partially optimized
as described above, but the optimization process will continue as better
stationary phases, detectors, and computer hardware and software systems
are developed.

4.2.4  Temperature Dependence

The retention time of a sample vapor in a GC column is a strong function of
the column temperature.  For this reason, conventional chromatographs
typically maintain an elevated column temperature constant to within 1°C
or better, or they utilize temperature programming to facilitate certain
separations.  In the portable instrument there is insufficient battery
power to run a controlled-temperature oven, so the capillary column oper-
ates at approximately room temperature.  Since room temperature varies
roughly between 0 and 45°C, the computer must be able to predict the re-
tention times of all peaks of interest over this temperature range.
                                    40

-------
The temperature dependence of retention follows directly from the ex-
ponential relation of equilbria in phase transformations such as vapori-
zation to which the partitioning of a solute between gas and liquid
phases is closely related.  The equilibrium partition coefficient K is
given by
                             K = C exp
                                         RT /

where

     C is a constant

     AH  is the heat of solution
       s
     R is the gas constant

     T is the absolute temperature

The solution process is usually exothermic (AHS is negative), thus increas-
ing T decreases K.

Over the limited temperature range at which the column operates, AHS is
assumed to be constant and the partition ratio k is found from Equation
(7) to be of the form
The constants A and AHS can be determined experimentally for each gas on
a miniature capillary column by performing separations over the tempera-
ture range and fitting an exponential curve to the data.  Figure 4.10 shows
the results of such a procedure for chloroform and methylene chloride on
an OV-101 capillary column.

To predict the retention time of a given gas, the computer only has to
measure the column temperature, insert the appropriate values of A and AHS,
and determine the partition ratio k.  Then using the air peak retention
time (t]fl) from the previous chromatogram, the gas retention time is calcu-
lated as
                             tR = tM (1 + k)
4.3  DETECTOR PERFORMANCE
A metal film thermal conductivity detector has been chosen for use in the
integrated gas chromatograph.  This device was chosen for its compatibility

                                   41

-------
              1.0

              0.8


              0.6
            -r 0.4
            z
            o
              0.2
CHLOROFORM
                            METHYLENE CHLORIDE
                3.21   3.23  3.25   3.27  3.29  3.31   3.33   3.35
               (39°C)                                  (26°C)
                       ABSOLUTE TEMPERATURE'1 x 1Q'3


Figure 4.10.  PARTITION RATIO AS A FUNCTION OF 1/T FOR  CHLOROFORM AND
              METHYLENE CHLORIDE.
with the miniature system, ease in fabrication, generality of  response,
and ruggedness.  In the many months of testing it has been shown  to  be  a
consistent performer.

4.3.1  Comparison with Other Detectors

There are a number of different types of detectors used in gas chromato-
graphy.  They measure one of a wide variety of physical parameters of the
output gas stream.  Of the detectors with relatively general response,  the
most commonly used are flame ionization detectors and thermal  conductivity
detectors.  Lately photoionization detectors have been shown to be a
general purpose detector with good analytical performance.  The ionization
detectors are typically known for their high sensitivity to hydrocarbons,
and the thermal conductivity detectors are used for applications  where
their generality of response to nonhydrocarbons is important.

The flame ionization detector works by introducing hydrogen in the output
gas stream of the chromatograph.  The resulting gas mixture is then  ignited
                                    42

-------
in a burner, the hydrocarbons  in. the output gas stream are burned, and
a small percentage of the combustion products become ionized.  These ions
are collected by electrodes and the current between the electrodes is
integrated to determine the mass of sample in an output peak.  The ability
of a flame ionization detector to measure small concentrations is due to
its low background current and the ability of electronic systems to mea-
sure small currents.

The most important property of a detector for many applications is its
detectivity.  Since ionization detectors are basically mass sensitive
instruments, their response is generally given in coulombs/gram and mini-
mum detectable quantities in grams.  Thermal conductivity detectors are
concentration sensitive devices where the minimum detectable sample is
usually given in parts-per-million.  The miniature gas chromatograph has
some unusual properties due to its small size and small gas flows.  It
places stringent requirements on the detector.  One aspect of these re-
quirements is the small sample that the detector system will be required
to measure.  In order not to overload the capillary column, the injected
sample must be kept very small, in most cases about 10 nl of gas.  For
many substances, it is desired to measure concentrations below the 5 ppm
level.  For gases with molecular weights around pentane, 5 ppm of 10 nl
of injected gas represents only about 150 x 10~15 grams (150 fg).  Because
of the gas flow rate of about 4-yl/s out of the column and the narrow peak
widths of less than 100 ms, the absolute maximum detector volume cannot
exceed 20 nl, without serious peak broadening due to dilution in the
detectgr volume.  Thus for a column with chromatographic efficiency simi-
lar to that of the prototype, the maximum detector dimensions are 100 ym
x 200 ym x 1000 ym.  If a shorter column is used or if more efficient
linings are developed, the maximum allowable detector volume would be re-
duced by a corresponding factor.

For purposes of comparison, the detectivity of a high quality flame ioni-
zation detector is 18 pg [6]  of, in this case, propane, some two orders of
magnitude larger than needed for the integrated gas chromatograph.  Even a
photoionization detector, which is reported [7J  to have significantly better
detectivity than the flame, can see no less than 6 pg of benzene, again
almost two orders of magnitude larger than is needed.  These figures were
found by injecting quite large volume samples into short, packed columns.
In the photoionization case,  5 ml of 0.3 ppb benzene was used.  Even
though the concentration of this sample is very low, the total mass of the
sample is still quite high, and the volume of injected sample is equal to
the total carrier gas flow for the miniature system for 25 minutes.  These
detectivity problems, coupled with the difficulties associated with trying
to fabricate an ionization detector in a region a few hundred micrometers
across, have limited the feasibility of making a miniature ionization de-
tector.

Conventional thermal conductivity detectors have not been known for their
good detectivities.  A Varian 3700 TCD can achieve sample detectivities
of 1 ng [8], about four orders of magnitude larger than needed for the
miniature system.  The Varian system also has a minimum detectable peak
concentration of about 200 ppb at the detector.   The important property of
thermal conductivity detectors is that they are concentration sensitive


                                    43

-------
detectors, that Is, the output signal is proportional to the concentration
of sample gas in the output gas stream.  If the dimensions of the detector
are reduced, the sensitivity of the detector is not greatly affected.
This allows a thermal conductivity detector to be made on the size scale
required for the integrated GC without a penalty for the small dimensions
involved.  The detector retains the concentration performance of the larger
detectors, and since the detector volume is so much smaller, the mass
detectivity is reduced to very small levels.  The integrated TCD described
below is capable of detecting below 150 fg of injected pentane and has a
detector volume of less than 1 nl.

4.3.2  Detector Operation

The nickel film integrated TCD used with the miniature gas chromatograph
is shown again in Figure 4.11.  A nickel film  was chosen as the sensing
element because of its relatively high temperature coefficient of resis-
tivity (TCR), moderately high melting point, ease of deposition and
selective removal, and relative inertness.  Most larger TCD's use tungsten
or tungsten alloys for the sensing element which have about the same TCR,
a higher melting point, but more difficult deposition than nickel.  The
structure and fabrication sequence used for the detectors can accommodate
a wide variety of metals as sensors; nickel has performed adequately in
the initial tests.

There are a number of ways to heat the sensor.  In a self-heated system
the choices are among the electrical circuit configurations used to drive
the detector.  The three most common are the bridge, constant current, or
constant temperature.  Each different drive configuration produces dif-
ferent sensitivity, thermal time constants, temperature extremes, linearity
                            PYREX GLASS
                            Si GC SUBSTRATE

                          •ETCHED GAS CHANNEL
                          1000 A* NICKEL FILM
                      ETCHED  DETECTOR CAVITY
1.5 PYREX
 GLASS
DETECTOR Si
SUBSTRATE
                  Figure 4.11.  DETECTOR CROSS SECTION.
                                    44

-------
and detectivity.   The most important  parameter  for  the miniature system
in environmental monitoring applications  is  detectivity.  Assuming that
the detector noise remains constant,  the  electronic system with the
highest sensitivity will be optimum for the  miniature system.  Figure
4.12 shows a graph of calculated detector sensitivities as a function of
temperature for the bridge, constant  current, and constant temperature
configurations.  The choice is basically  between constant current and
temperature.  The maximum temperature limit  for these detectors is about
300°C above ambient, but they are usually operated  below this point.  The
constant current mode results in the  highest sensitivities at higher tem-
peratures due to a thermal feedback effect because  of the positive TCR of
the nickel resistors.  At a given operating  point a sample peak causes
the thermal conductivity of the gas to decrease.  This causes an increase
           1000
         UJ
         o
         0.
         
-------
in the detector temperature which causes an increase in resistance and
subsequently an increase in power, l2R(T), dissipated in the device.
This increase in power also increases the detector temperature which again
increases sensitivity.  This feedback effect is normally stable, unless
the operating point temperature is set too high.  If the current source
can put out unlimited power, then the detector can enter so-called thermal
runaway and destroy itself.  Normally the characteristics of the current
source are designed to limit the maximum power available to drive the
detector so this condition is averted.  This thermal feedback can cause
nonlinear response for very large peaks as described below.

The constant temperature mode can also be used, where the detector voltage
is constantly adjusted to keep the detector at a particular temperature.
When a gas peak goes past the detector, the voltage drive is reduced,
reducing the power dissipated by the detector.  .This mode is not as sensi-
tive as constant current at moderate (100-200°C) tempeatures, but does
prevent thermal runaway.

Basic thermal and electrical theory can be used to calculate the operating
point, temperature, temperature excursions, sensitivity, and linearity of
the integrated TCD.  It is most convenient to assume that the structure
is always in thermal equilibrium.  The thermal time constant of the device
is about 1 ms, much smaller than any peak width, so that the input power
to the device always matches the power dissipated by the detector, which
is mostly conducted through the carrier gas.  In the constant current case
the dissipated power is simply l2R(T).  The power dissipated by the detec-
tor is approximately GQT(1 + 3T/2) where Go is the thermal conductance of
the combination of carrier gas and detector support and 3 is the tempera-
ture coefficient of thermal conduction of helim.  To solve for the detec-
tor temperature a form of R(T) must be found.  For the nickel films used
in these detectors the resistance is quite closely given by
                            R(T) = R  exp(aT)                         (8)
where R0 is the detector resistance at ambient temperature, and T is the
temperature above ambient.  For nickel films the TCR, a = 0.00426 °C~1,
is appropriate to use.  To solve for the operating point temperature the
equation
                                                                      (9)
must be solved.  This is usually done numerically, but a graphical solution
may be employed for purposes of discussion.  The graph shown in Figure
4.13 shows the power dissipated by four constant currents through the
detector for a range of detector temperatures to 280°C above ambient.
Also plotted is the power dissipated by the detector for those same tempera-
tures.  The points of intersection, A, B, and C, are the operating point
temperatures for 6, 8, and 10 mA.  For the conditions used in this graph,


                                    46

-------
              0.1

          o
          QL
          a:
          o
          o
          UJ
          h-
          UJ
          o
             0.01
            0.001
                0  20 40 60 80 100   140    180   220    260

                      DETECTOR TEMP. ABOVE AMBIENT, °C
Figure 4.13.
              GRAPHICAL SOLUTION TO DETECTOR OPERATING POINTS  FOR DIF-
              FERENT DETECTOR CURRENTS.   R  = 200 ft,  G  = 0.002  °C/Watt.
                                          o           o
R0 = 200
           and GQ = 2 x 10~3 W/°C,  the detector goes into thermal runaway
before 12 mA is reached.   This first order theory is sufficient for analyses
of operation below about  200°C , above ambient.   In fact,  the device does
not run away at 12 mA in  the above sample because the TCR of the nickel
resistor decreases enough at high temperature  to remain  in a stable state.

Two of the more important characteristics of the detector are sensitivity
and linearity.  The small signal sensitivities are determined by differ-
entiating equations such  as (9)  to get dV/dC,  the change in output voltage
for a small change in sample gas concentration.   This is usually done as-
suming that dG/dC, the change in thermal conductance for a given change in
concentration, is constant.  These assumptions are generally correct for
                                   47

-------
gas peaks, but for air peaks the voltage from the detector can double,
which can introduce considerable nonlinearities.  These nonlinearities
in response to large peaks can be computer corrected if a form for the
response is known.

In order to find the output voltage of the detector for a given peak con-
centration, Equation (9) is solved twice, once for the operating tempera-
ture TQ, and again for the peak temperature T' when the thermal conductance
is reduced by a gas peak to Go.  The output voltage is then
                          V  = IR(T') - IR(T )


The output voltages for a given change in G0 have been found for a set of
detector currents I and the results are shown in Figure 4.14.  Here current
is varied as a parameter and the operating point temperature and the out-
put peak voltage are both found.  The experimental points are determined
by measuring the peak voltage for the same set of currents, and measuring
the temperature of the detector by measuring its resistance.  The agree-
ment between the points is excellent over a wide range of operating con-
ditions.  A wide variety of curves like this can be determined for in-
jected peaks of different sizes.

To determine the linearity of the detector, the output voltage must be
found as a function of changes in thermal conductivity, for a fixed de-
tector current.  In order to convert these results to output voltage as
a function of mole fraction of the sample gas in the carrier gas, the
thermal conductance of the gas mixtures must be known.  Theoretical re-
sults are not sufficiently accurate for most gas mixtures, so experimental
results of thermal conductivities of gas mixtures as compiled by Tseden-
berg [9J were used.  The experimental points for argon in helium were
curve fitted as shown in Figure 4.15.  This result was used to find the
thermal conductivity change for a given argon concentration, which was
then used to find the output voltage at a given detector current.  The
output voltage is then divided by the argon concentration to produce the
output detector sensitivity.  A graph of theoretical detector sensitivity
vs argon mole fraction is shown in Figure 4.16 for a detector current of
12 mA and an operating point temperature of 50°C above ambient.  The
linearity is good over the range from a part-per-million, which is near
the lower limit of detectivity of the detector, to a few percent argon in
helium, which is near the maximum sample concentration at the detector
during the passage of the airpeak.

The mounting of the detector in the output gas stream introduces sources
of noise to the system such as mass flow artifacts and flow noise.  The
detector operates by dissipating heat by conduction through the gas stream.
Since the gas is moving past the detector, there is an additional com-
ponent in the detector heat balance equation given approximately by
CvVT,j/2, where Cv is the specific heat of the gas, V is the column flow
rate, and T^ is the detector temperature above room temperature.  At a
constant carrier gas flow this heat loss is incorporated in Equation (9)


                                    48

-------
            10,000
          >
           E
          UJ
          ID
           111
           Q.
           UJ
           O
           O
           o:
             1,000
               100
                10
                      A •
                     I = 17ma
                                            '1 = 16
                                         1=15
                                      = 14
                               1 = 12
'1=10
                        1 = 8
                       = 6ma
                                          • EXPERIMENT
                                          A THEORY
                 "0   20   40   60   80   100  120  140
           DETECTOR OPERATING POINT TEMP ABOVE AMBIENT, °C
Figure 4.14.  DETECTOR OPERATING POINT TEMPERATURE AND NITROGEN PEAK VOLT-
              AGE AS A FUNCTION OF DETECTOR CURRENT.
in the Go term, but if the magnitude of the gas flow changes, the detector
temperature and hence the output signal will change.  The flow can change
due to either changes in carrier gas supply pressure or at the introduction
of higher pressure sample gas at the injection valve.  The injection arti-
fact can be a rather large signal, at high gain, but is over by the time
the air peak reaches the detector.  An example of such a signal can just
be seen, immediately after injection, in Figure 4.4.  The change in signal
with carrier gas pressure is noticeable in the prototype instrument, but
does not degrade performance.  The change in carrier pressure is due to
the on/off valve and surge tank arrangement in the miniature carrier gas
supply system.  The pressure in the surge tank drops about 7 x 10  Pa
(0.1 psi) during a chromatogram.   That drop introduces a drift in the
                                   49

-------
       0.12
>-
(-;
>
\-
o
_/
Q
0 «
8-w
_i
(J
Q
1
E
       0.08
a.
LU
       0.04
           0
      20         40        60        80
            PERCENT ARGON IN HELIUM
                                    100
  Figure 4.15.  THERMAL CONDUCTIVITY OF Ar/He GAS MIXTURES AS A FUNCTION  OF
               THE PERCENTAGE OF Ar IN THE MIXTURE.
           1.6

         2 1.5
         o

         1"
         LL
         LU 1.3
         O
         LU
         Q.
           1.1
         _
         O
         >  1
            106
     105
104
id3
10
 ,-2
                             ARGON MOLE FRACTION
Id1
  Figure 4.16.
LINEARITY OF DETECTOR SENSITIVITY AS A FUNCTION OF ARGON
MOLE FRACTION IN THE HELIUM OUTPUT STREAM.
                                    50

-------
detector signal approximately equivalent to the drift introduced by am-
bient temperature variations.  A filtering algorithm in the computer
program is used to compensate for these drifts.

The flow of gas past the detector also causes some degree of noise.  This
noise is thought to be due to extremely slight flow instabilities in the
detector gas channel.  The flow itself is highly laminar, due to the small
dimensions and low gas velocity, but these slight instabilities can be
detected as noise.  In the prototype instrument electronics, this detector
noise is about the 0.5-yV peak-to-peak level, somewhat under the 2-yV
peak-to—peak noise of the current source and input noise of the detector
amplifier.  At high detector currents, which produces higher detector
sensitivity, the flow noise becomes the dominant noise source of the sys-
tem.

4.4  COMPUTER SYSTEM

The miniature gas chromatograph described in this report is capable of
significant analytical performance.  The unique capabilities of the device
are realized only when the device is controlled by, and the results ana-
lyzed by, a sophisticated computer system.  Since the chromatograph can
separate and detect up to 10 vapors in less than 10 seconds, the data ana-
lysis system must operate quickly and be compact enough for inclusion in
a portable package.  In addition, in order to make a truly useful instru-
ment, the operation, calibration, data storage and display functions must
be convenient and complete.  The computer hardware system and its asso-
ciated program have been designed to fulfill these requirements as com-
pletely as possible given the present state of the art in miniature, low
power electronics.

The computer system was designed to be as flexible as possible, giving the
computer access via the A/D converter to the pressure and temperature
sensors and the amplifier outputs which describe the state of the analyti-
cal system.  The computer has been programmed to use this information to
run the system and has a complex data analysis section to process the
detector output signal.  In addition, a significant portion of the com-
puter program is dedicated to interfacing the electronics system to the
user via a keyboard and display.  The algorithms and programs necessary
to run the system are described in the remainder of this section.

4.4.1  Data Analysis

The most difficult part of the computer program is associated with analyz-
ing the detector output signal.  A typical chromatogram is a rather complex
signal, especially for a modest computer system.  There is often a pres-
sure artifact after injection, a very large air peak, followed by any num-
ber of potentially very small gas peaks.  The uncontrolled GC temperature
introduces considerable drift to the amplifier output along with both
flow noise and noise from the electronics.  In order to measure the con-
centration of an output peak the area of the output peak must be measured
and compared to the area of the air peak.  The major analytical problem
is thus reduced to finding the area of a small output peak in the presence
of noise and drift.

                                    51

-------
There are a wide variety of methods to detect the presence of small signals
in noise.  Given some knowledge of the properties of the signal and the
noise, it is usually possible to select a system which enhances the ability
to measure the desired attribute of the signal and to minimize the effects
of the noise on that measurement.  In the case of the gas chromatograph,
the signal is the detector voltage, which is proportional to the concen-
tration of gas in the output gas stream, and the noise consists basically
of random fluctuations in detector output, electrical noise introduced by
the amplifier, and a variety of drift components.  Due to the chromato-
graphic process, the gas concentration and hence the detector signal,
x(t), is quite nearly gaussian shaped with respect to time
This gaussian peak can be described as having three distinguishing charac-
teristics, a retention time, to, a peak area, Ax, and a standard devia-
tion, ax, and all three must be measured by the computer system.

Filtering Algorithm —
In order to perform these measurements, the analytical problem is broken
into two parts, a filtering procedure to maximize the signal to noise (S/N)
ratio, and an analysis of that filtered data.  The filtering procedure was
chosen to maximize the ability to determine the existence of, and measure
the retention time of, the gaussian peak.  Filtering theory is then used
to find the optimum way of finding the retention times, t0.  The transfer
function H(u), of an optimum filter is found to be [lO]
                                     exp <-j2m)to)                   (11)
where
      *
     X (u) is the complex conjugate of the  input Fourier transform

     Y (u) is the power spectral density of the noise

     K is a constant

In order to solve for the transfer function of an optimum filter, some form
of the power spectral density of the noise must be assumed.  For this pro-
totype the noise was assumed to be white, that is YJJ(U) is a constant.  For
white noise the optimum filter will have the same shape as the input signal,
so that in the case of a gaussian input signal the output of the filter
would also be a guassian.  The standard deviation of the output signal
(0out) would be given by
o    -f<
 out   •<{
                                     2     2
                                     .  + (?„
                                     in    F
                                    52

-------
where a^n is the standard deviation of  the  input gaussian signal and Qp is
the standard deviation of the filter in the time domain.

In order to simplify the determination  of the  location  of the signal maxima
and points of inflection, a differentiator  was included in the filtering
algorithm.  Differentiation at this stage is tolerable  since much of the
noise has been removed by the optimum filter.

The overall impulse response of the optimum filter and  the differentiator
is given by
                       h(t)
  3 fc exp r72
a         \  2a
 X        \    X,
                                          (12)
This filtering function is graphed as  a  function  of  time in Figure 4.17.
When the original gaussian peak is operated  on by this filtering function,
a different output is obtained which is  given by  the convolutional integral
                      Y(t)
=/:
X(T) h (t - T)  dT
(13)
Using Equations (10),  (12),  and (13)  the  filter  output  signal Y(t), is
given by
                                 -h(t)
                                                             TIME
      Figure 4.17.   IMPULSE RESPONSE,  h(t), OF THE OPTIMUM FILTER.
                                   53

-------
                  Y(t)
                         VH
                        /2rF
               3 "o -
           (t - V
              2 a2
where
      y      x
     A^ is the area of the input gaussian

     A^ is a constant from Equation (12)

     a  is the input gaussian standard deviation
      X

This function is shown in Figure 4.18.  This is the filtered gaussian wave-
form which must be measured to determine the concentration of the gas peak.
In order to do that, the original standard deviation, ax, and the original
amplitude must be reconstructed from measuring this waveform.  The stan-
dard deviation can be determined from tmax and !„-,•„, since it can be shown
from Equation (12) that tmin - *"    = °~
                                                                      (15)
                                 Lmin»
Furthermore, the area under the positive portion of Y(t) is
             /:<
Y(t) dt =
                                           /2iF
          Y(t)
                            lmax
                                           TIME
Figure 4.18.
THE RESULTANT WAVEFORM AFTER FILTERING A GAUSSIAN INPUT
SIGNAL.
                                    54

-------
so that the original peak area, AX, can be recovered by performing the

calculation.


                                A  = KS 0                            (16)
where K is a constant which is the same for all input gaussians.  Thus by
measuring the area S and tmax and tmin, the gaussian area can be found.
This method is theoretically more reproducible than measuring the area of
the original gaussian peak, since much of the noise which could interfere
with the integration has been removed by the filtering process.

Filter Realization —
In a digital system, the filtered output signal, Y(t), cannot be deter-
mined continuously, but must be calculated at discrete time intervals.
This is achieved by first sampling the input waveform by an A/D converter
to obtain a number proportional to the input voltage.  As the input is
repeatedly sampled, a set of numbers is obtained as a function of time.
This set is multiplied by a set of filter coefficients which are found by
sampling the impulse response of the filter, h(t) at the same rate.  Each
filtered point is the sum of a large number of sampled points multiplied
by the filter coefficients.  The larger the number of multiplications
performed, the closer the resulting sum will be to the actual value of
Y(t).

In a microcomputer realization of the digital filter, the speed at which
the multiplications can be performed is the limiting factor.  A compromise
must be found in the chosen sampling rate of the analog signal, the pre-
cision of the numerical representation of the signal, the precision of
the filter coefficients, the number of multiplications performed for each
sample point, and the way the data points are stored in the computer's
memory.

In order to perform the filtering function, the convolutional integral,
Equation (13), must be approximated in discrete form as
                                      Xi\-i
where
     Y  are the filtered sample points
      Jv

     X. are the sampled input data points

     h  are the filter coefficients

The computer program to perform this filtering function is described below.

The digital representation of the detector amplifier signal is stored in
the computer memory along with the last 63 such sample points.  After each
sample is taken and stored, the accumulator is zeroed, every other point

                                    55

-------
is retreived from memory, multiplied by the correct filter coefficient for
that point, and the result is added to the accumulator.   After 31 such
multiplications the number in the accumulator is the filtered result for
that set of 64 data points and this result is stored.   About 3 milliseconds
later another sample command is given, the 64th data point is erased
from memory, all previous data points are shifted down one place, and the
new data point is stored to begin the process again. This process is shown
in graphical form in Figure 4.19.  Here the filter coefficients are shown;
these are the multipliers of the delayed sample points mentioned above.
When the gaussian peak is processed by the filter, the result is shown
below.  The filter is seen to produce the signal, Y(t),  which is nearly
the derivative of the input gaussian.  The retention time of the peak is
the zero crossing point of the output signal, and twice the standard
deviation, ay, is the positive peak to negative peak time.   In order to
obtain A^, the area of the original gaussian, the area S must be found.
This is done by a simple point by point numerical integration of the
filtered signal.
                               x(t)
                                • • .
          TIME
                                          IJLu
                                         GAUSSIAN
                                         INPUT
POINT BY POINT
MULTIPLICATION
*••.
*.!
h(t)
1
<
• •
• •
- '-.;
'..
•
                                                       FILTER
                                                       COEFFICIENTS
         SUMMATION          I.
                     	•*
                    Y(t)
          PREVIOUS POINTS
                                                       OUTPUT
                                              	jxx-DATA
                                              
-------
Errors—
In the miniature system the main sources of error in the peak area calcu-
lations are due to the finite sampling rate, which limits the precision
in finding Oy, if ay is calculated directly from tm^n and tmax.  Since
there are usually only about 30 sample points between tm^n and tmax, an
uncertainty of one sample point is about a 3% error.  In the presence of
noise, the uncertainty is increased due to nonideal waveforms.  There are
basically three ways to improve this performance.  Conceptually the eariest
is to sample more frequently, which would increase the number of points
between tm±n and tmax, and thus improve the resolution of determining oy-
In order to achieve large increases in performance, however, the compu-
tational speed would have to be proportionally increased.  The micro-
processors necessary to perform these calculations are not presently
available, so this solution is not now feasible.  The other two methods
use interpolating filters or curve fitting to achieve effectively greater
time resolution than the sampling interval.  The curve fitting procedure
was found to be very helpful, but is very time consuming, taking many
seconds per peak area calculation.  This is not desirable for the portable
instrument.  It is also possible to use a class of interpolating filters
to increase the effective time resolution.  This procedure seems to be
the most promising to increase the accuracy of the ay determination and
thus the area measurement, and it will be investigated at the earliest
opportunity.

An additional source of error is due to the action of the differentiator
in the filtering algorithm on the input signal.  In the presence of drift,
the differentiator will change the drift into an offset.  This offset
shifts the position of the zero crossing of the filter output signal.  For
peaks with large retention times and low amplitude, the baseline offset
signal may be so large that no zero crossing is detected.  The computer
program will be modified in the future to compensate for this baseline
offset to increase ability to detect the later peaks.

Filter Simulation—
The effectiveness of the filtering system can be seen in the computer
simulation results shown in Figure 4.20.  The top output shows a gaussian
peak plus 1/f noise before the filtering,and the lower output is after the
filtering algorithm.  A computer simulation is useful in analyzing the
performance of the filter since the amplitudes of the peak and the noise
can be easily adjusted and the "retention time" of the gaussian peak is
exactly known.  The filter parameters and standard deviations can also
be changed to characterize the filter over a wide range of conditions.  A
systematic characterization of various filtering and area algorithms will
be the subject of a forthcoming report.

Peak Finding—
The filtering scheme mentioned is designed to increase the signal to noise
ratio of the detector output signal and thus make it easier for the com-
puter to locate and measure the peaks.  The filtering process is a part of
a section of the program dealing with finding and measuring the peaks.  A
simplified flow chart of this program is shown in Figure 4.21.  The main
criteria for locating a peak is the steep zero crossing in the filtered
data.  If a positive maximum above a threshold is found in a time window,
followed by a zero crossing, a peak is said to occur.  It is not until the
zero crossing, however, that the computer has determined that a peak exists,

                                    57

-------
 250.00
  0.00_
-250. 001	
    0.00
                     1.00
                        I
                                      2.00
                                                       3.00
                                     TIME < MO >
 250.00.	
-250. HBI...  . i .   i     i    I    i    i     i    I    i    i     i
Figure 4.20.
COMPUTER SIMULATION  OF GAUSSIAN PEAK WITH  NOISE.   (a)   Before

filtering;  and  (b) after filtering.
                                       58

-------
                      PEAK ANAYLSIS  FLOW
                                  i
                             SAMPLE  DATA
                              AND FILTER
            GET RET.  TIME
            OF NEXT  PEAK
                                  END
                             CHROMATOGRAM
INTEGRATE PTS &
DETERMINE SIGMA
                                    YES
                          AREA CALCULATIONS
                             AND  DISPLAY
     Figure 4.21.  FLOW CHART FOR PEAK FINDING AND AREA CALCULATIONS.
and at that  point half of the filtered peak data has already been calcu-
lated.  The  computer thus needs to continuously store the filtered output
points so that  the data is available when a legitimate zero crossing is
found.  After the zero crossing, the second half of the peak data is also
stored, and  the complete peak data is later analyzed by the peak area calcu-
lation program.

Swept Time—
In order to  achieve maximum benefit from the filter, the standard deviation
of the filter should be very close to that of the input gaussian peak.   In
a chromatogram  this is very rarely the case.   The air peak is quite narrow,
with a standard deviation of about 20 ms.   For a peak with a long retention
time, the standard deviation can increase by a factor of five or more.   Such
                                  59

-------
a wide peak would not be matched to the characteristics of the filter and
the benefits of the  filtering would be substantially reduced.  It is not
practical to change  the filter characteristics  during the course of the
chromatogram,  so "time" is changed instead.   The  sampling period is
linearly increased during the chromatogram so that  each peak is a constant
number of sampled points wide.  This then guarantees effective filtering
for all of the peaks at the slight disadvantage of  having to convert from
sample point units to real time for all retention time calculations.  This
process of changing  the sampling rate has been  called "swept time".  The
match between standard deviation increase and linearly increasing period
is quite close,  so that this swept time algorithm is well suited to the
needs of the portable chromatograph.

4.4.2  Operating Program Flow

The peak finding section is a part of a larger  program which operates the
system.  As shown in diagram form in Figure  4.22, the program proceeds as
follows.  Upon a signal to start an analysis the  computer first takes a
reading of the column temperature.  This is  used  mainly to calculate the
expected retention times of the sample peaks of interest for that chroma-
togram.  Time windows are then set up so that the computer knows where to
look for the zero crossing of a peak.  The high gain amplifier is zeroed
to take care of any  amplifier zero drift since  the  last analysis and the
high pressure valve  in the carrier gas supply is  opened to charge up the
   BEFORE
 INJECTION

 POWER UP

 MEASURE TEMP.

 CALC.  RET.  TIMES

 SET  He PRESSURE

 ZERO AMPLIFIER

 ACTIVATE PUMP

 CHECK PRESSURE

 PULSE VALVE
     DURING
  SEPARATION
       AFTER
  CHROMATOGRAM
FILTER  IX OUTPUT
DETECT  AIR PEAK
INTEGRATE PEAK
STORE  AIR DATA
FILTER  5000X OUTPUT
DETECT  GAS PEAKS
INTEGRATE PEAKS
STORE  GAS DATA
CALC. PEAK AREAS
DETERMINE CONC.
CHECK  FOR ALARM
UPDATE  TWA DATA
DISPLAY?
POWER  DOWN
        Figure  4.22.  SIMPLIFIED FLOW CHART OF  THE MAIN PROGRAM.

                                  60

-------
pressure regulator surge tank to the correct pressure.  The computer
monitors the pressure sensor on the surge tank until the desired pressure
is reached, when it closes the high pressure valve.  The computer then
signals the sample pump to withdraw, pulling in an atmospheric sample
through the sample input system.  By monitoring the pressure drawn by the
sample pump, the computer determines when the pump has finished drawing
in the sample.  The pump is then reversed, pressurizing the sample.  When
the desired sample pressure is reached, the computer fires the injection
valve, which introduces the sample to the head of the separating column.
The computer then starts sampling the detector output through the A/D
converter.  Just before the air peak is expected, the swept time circuitry
is activated, which begins to lengthen the signal sampling period.  After
the air peak is detected the computer switches to the high gain amplifier
output and starts searching for peaks.  When a timer expires signaling the
end of the chromatogram, the computer computes the area of the peaks,
calculates the respective concentrations, updates the appropriate maximum
and TWA concentration registers, and then shuts itself off, awaiting a
command from the keyboard or the automatic timer.

Calibration Routine—
It is possible for this machine to perform an automatic calibration cycle.
It has been programmed to automatically find and measure the retention time
of a gas peak and store the retention time and amplitude calibration factors
in memory.  To calibrate the machine for a particular gas, that one gas is
connected jto the input of the instrument and the calibration routine is
selected.  The temperature coefficient for that gas must be previously
entered, along with the known concentration of the calibration gas.  The
computer then performs the analysis and updates all pertinent calibration
coefficients.

This simple routine should reduce the amount of periodic service of the
device for calibration purposes and allows for on-site calibration to in-
sure the accuracy of the portable instrument.

Keyboard and Display Program—
For the prototypes, a simple keyboard and display were desired.  The amount
of front panel space is limited so a large keyboard is not practical.  A
16-switch keyboard was chosen along with a 4-digit display.  When coupled
to the computer these limited devices are capable of a number of functions.

In the prototype instrument a large number of the internal parameters are
capable of being changed or updated by the user.  There are 10 system
parameters plus 9 gas parameters for each of 10 different gases which can
be displayed or changed by the user.  In addition there is a calibration
mode, normal sampling mode, and automatic sampling mode which can be
selected.  A simplified chart of the keyboard routine is shown in Figure
4.23.  When the machine is in its quiescent state, it is waiting for a
function key, either A, B, C, D, or E.  If A is pushed the machine begins
a new analysis.  B is used for displaying and changing the system param-
eters.  For example, if B and then 1 are pushed, system parameter 1, the
gas peak threshold, is displayed.  C is used to change the value of a gas
parameter.  A sequence here might be to change the temperature coefficient
of gas number 5.  The button sequence would be C, 5, and then 3, the button
3 for the temperature parameter.  The display would then zero, waiting for
the new value to be entered.  To input  .00029, for example, the button

                                    61

-------
COMMAND      FUNCTION                      DESCRIPTION

   A        Survey mode         The machine performs one sample analysis
                                and updates concentrations of programmed
                                gases.
   B        System param-       To display a system parameter the param-
            eter display/       eter is entered.  After the parameter is
            change              displayed, pressing 3 will allow the user
                                to change that parameter.
   C        Gas parameter       To change a gas parameter, the gas # and
            change              parameter # are entered, followed by the
                                new value for that parameter.
   D        Gas parameter       Entering the gas # followed by parameter
            display             // displays the value of that gas parameter.
   E        Sample calibra-     In this mode the machine determines the
            tion                overall concentration multiplier and re-
                                tention time "A parameter" of the gas
                                under calibration.  After choosing a # for
                                the gas, the known concentration of the
                                calibration gas is entered.  The machine
                                then performs one analysis and updates
                                the necessary parameters.
Automatic Mode:  The machine samples the atmosphere and updates maximum
                 concentration and TWAs of programmed gases with the
                 sampling interval specified by system parameter 7.

           Figure 4.23.  KEYBOARD COMMAND DESCRIPTION.
sequence F,0,0,0,2,9,F is used, the first F is interpreted as a decimal
point and the last F is an enter button to tell the machine that the num-
ber is complete.  The D button is similar to the B, but for display of
the gas parameters.  The sequence D15, for example, displays the concen-
tration of gas 1.  The last button, E, activates the calibration routine;
the sequence E 4 1000 calibrates gas 4 at 1000 ppm.  A complete list of
the function keys and gas and system parameters is listed in Figure 4.24.
Mistakes are cleared by simply pressing another function key instead of
the "enter" button.  The 4-digit display can show up to a 10-digit number
by shift keys which move the display one place to the right or left.
Since many of the displayed numbers are betwen 9999 and 0.001, this shift
is not always required.

Power Down—
The present digital electronics system uses a Z-80A microprocessor, six
EPROM's, four RAM's, plus a number of support chips.  During operation
the device consumes about 300 mA at 5V, or 1.5 W.  In order to save power
a novel power down system for the digital electronics is used.  The general
scheme is to power the microprocessor only during those times that is
actually computing, and to power down whenever it is waiting for a key-
board entry.  This system also works in between key strokes from the

                                   62

-------
                             GAS  PARAMETERS
     Parameter Number                        Definition
             0                       Select/deselect  flag
             1                       Retention  time "A parameter"
             2                       Retention  time "B parameter"
             3                       Overall  concentration calibration
                                    multiplier
             4                       Overall  concentration correction
                                    with temperature
             5                       Concentration
             6                       Time weighted average concentration
             7                       Maximum  concentration detected
             8                       Alarm level
             9                       Reserved

                            SYSTEM PARAMETERS
     Parameter  Number                        Definition
             0                       Maximum  chromatogram length
             1                       Gas detection threshold
             2                       Carrier  pressure
             3                       Valve injection  pressure
             4                       Thermistor constant V,
             5                       Thermistor constant 3
             6                       Thermistor constant l/T
             7                       Automatic  mode wait timer
             8                       Swept time t^>
             9                       Swept time I/a
             Figure 4.24.  GAS AND SYSTEM PARAMETER LISTING.

keyboard.  For  example, when the computer finishes a chromatogram, it
stores the appropriate concentrations in nonvolatile CMOS memory and then
turns off power to itself.  When a concentration is  requested from the
keyboard, the D button is pushed, which produces a signal which turns on
power to the microprocessor.  The computer recognizes the button, goes to
the display part of the program and then turns power off again.  The
entire button pushing sequence takes about 1 ms, which is the amount of
time the microprocessor draws power.  As each additional button is pressed,
the machine continues to immediately power down.  After the desired

                                   63

-------
concentration Is displayed, the computer again powers down;  the number is
retained in the display driver and continues to be displayed.   This power
down system has allowed the battery size to be significantly reduced with-
out sacrificing instrument performance.

Future Improvements—
The computer system incorporated in this portable instrument represents
the most advanced possible system given the present state of electronic
development.  A number of improvements are foreseen using components which
are either announced for future delivery, or can be reasonably expected
to be available in the next year or two.  One such product is  a low power
Z-80A.  A product essentially equivalent and software compatible with the
present Z-80A but requiring only 20% of the power should be available in
1981.  Along with lower power memory and support circuits also soon to be
available, the battery required to drive the digital system can be further
reduced without reducing performance.

New digital signal processing devices are just being introduced.  These
devices are basically special purpose microprocessors, optimized for the
multiplications and divisions of high precision numbers which is most
important for signal processing applications.  A device such as these
could be used to prefilter the signal, freeing the microprocessor from
this job.  The microprocessor could then use a more elaborate peak detec-
tion program, and since it would not have to perform the filtering, the
sampling rate of the detector signal could be increased which would in-
crease accuracy.

There are some changes and additions to the signal processing algorithms
which should be included in future versions of the instrument.  The base-
line drift problem needs additional study to more reliably locate peaks
with long retention times.  In this prototype the problem of overlapped
peaks has not been addressed.  The filter used to reduce noise has been
designed to be optimum in the presence of white noise.  It appears that
the actual power spectral density of the noise decreases with frequency
proportionally to the reciprocal of the frequency, "1/f noise".  This
will require some changes in the filtering function to obtain an optimum
filter.

There are also improvements likely in the keyboard, display, and interface
circuitry which will increase the operator convenience and allow the in-
strument to automatically output the concentration data taken during a
day to a larger computer for further processing, storage, or hard copy
display.  A larger computer could also periodically monitor the performance
of the instrument, pointing out long-term changes in the calibration, or
troubleshooting the device in cases where repairs are needed.
                                    64

-------
                             5.  CONCLUSION
A prototype miniature gas analysis system has been designed and built
using a combination of new technologies.  The techniques of integrated
circuit processing have been utilized to miniaturize the components of a
gas chromatography system to a size which is compatible with a portable
package.  The electronics system used in the gas analyzer is made from
recently introduced, state-of-the-art microprocessor, memory, interface,
and analog circuits.  The combination of devices results in a portable
instrument with unique capabilities.

Some of the significant capabilities of the prototype instrument are:

     (1)  Automatic atmospheric analysis for 8 hours.

     (2)  Measurement of the concentration of 10 different toxic
          gases simultaneously.

     (3)  Concentration information within 30 seconds of sampling.

     (4)  Concentration measurements from the sub-10 ppm to the more
          than 1000 ppm level.

     (5)  Reproducibility within +5% at the 95% confidence level for
          several substances between 100 and 1000 ppm.

     (6)  Calculate and store the time-weighted-average and peak
          concentration for each of the gases.

     (7)  Convenient display of any concentration or system calibra-
          tion factor.

     (8)  Sound an alarm when any gas concentration exceeds a preset
          maximum value.

     (9)  Provide automatic calibration for any gas within the instru-
          ments measurement range.

    (10)  Relatively small size and weight for convenient portable
          operation.

The miniature gas chromatograph performs the actual separation and detec-
tion of the constituents of the gaseous sample.  It consists of a 1.5-m
long separating column, a sample injection valve, and a thermal conduc-
tivity detector all of which are integrated on a 5-cm diameter silicon


                                   65

-------
wafer.  The column is formed by etching a long spiral groove on one sur-
face of the silicon wafer and the groove is converted to a tube by hermeti-
cally sealing a Pyrex glass cover plate to the silicon surface.  The inte-
grated sample injection valve consists of a solenoid actuated diaphragm
valve in conjunction with an etched silicon valve seat and orifice.

The integrated thermal conductivity detector consists of a miniature nickel
film resistor supported in the output gas stream by a glass and silicon
structure.  All of the components of the chromatograph are fabricated using
novel "micromachining" techniques.

A small servo-driven syringe pump serves to draw in the air samples from
the environment and pressurize  them for injection into the column.  The
helium carrier gas for the chromatograph is supplied in replaceable 15-
cnr5, high pressure gas cartridges.  The gas is delivered by a miniature
pressure  regulator consisting of an integrated silicon valve, resistive
gas path, pressure sensor, and IS-cm^ surge tank.  Both the sample pump
and regulator are under the direct control of the computer.

An advanced computer system controls all aspects of the operation of the
instrument.  The system consists of a powerful microprocessor chip, pro-
gram memory, data memory, an analog-to-digital converter, a keyboard,
and display.  The computer controls the automatic sequence of events which
cause the instrument to periodically inject atmospheric samples to the GC.
As the separated sample peaks evolve from the chromatograph, the computer
filters the data to remove noise, measures their retention times for pur-
poses of identification, and determines the concentrations of the gases
in the atmospheric sample.  This data is used to update the peak and TWA
concentrations of the 10 selected contaminant gases.

The instrument is capable of operation in two different modes.  The device
can be used as a survey instrument, for rapid and accurate sampling under
command of an industrial hygienist.  In this mode the instrument is in-
structed to take an atmospheric sample.  After about 30 seconds the analysis
is complete, and concentration information is immediately available for
any of the 10 selected contaminants.  This rapid analysis should prove
useful in determining the location and extent of potentially harmful
conditions around a work place.  The instrument is also designed to per-
form automatic analyses at about 1.5 minute intervals.  The device can
then be used to record the TWA exposure and peak concentrations encountered
by a mobile worker during an 8-hour shift.  In this mode an alarm is
activated whenever the concentration of a gas exceeds a preset value,
warning the worker immediately of potentially dangerous situations.

Realization of the miniature GC gas analysis system is another example of
the benefits made possible by the constructive application of silicon inte-
grated circuit processing techniques to problems in diverse fields.  The
instrument should greatly expand the ability to monitor workers' exposures
to toxic gases and therefore aid in the reduction of such exposures, and
provide better epidemological data in future health studies.  In such an
application, the device should prove to be a useful analytical tool and
should have significant impact on the fields of industrial hygiene and
occupational safety.

                                   66

-------
                               REFERENCES
1.   Terry, S.C., J.H. Jerman, "A Feasibility Study of a Pocket-Sized Gas
     Chromatographic Air Analyzer," Stanford Electronics Laboratories,
     TR No. 77-027, Stanford University, Stanford, Calif., 1977.

2.   Terry, S.C., "A Gas Chromatography System Fabricated on a Silicon
     Wafer Using Integrated Circuit Technology," Ph.D. dissertation,
     Stanford Electronics Laboratories, TR No. 4603-1, Stanford Univer-
     sity, Stanford, Calif., 1975.

3.   Concus, P., "On the Liquid Film Remaining in a Draining Circular
     Cylindrical Vessel," J. Physical Chemistry, Vol. 74, 1970, p. 1818.

4.   Bouche, J., M.J. Verzele, Gas Chromatography, Vol. 6, 1968, p. 501.

5.   Golay, M.J.E., "Theory of Chromatography in Open and Coated Tubular
     Columns with Round and Rectangular Cross Sections," in Gas Chromato-
     graphy (Desty, ed.), Butterworths, London, 1958, p. 36.

6.   Dick, R., C.H. Hartman, "Critical Evaluation of the Response Charac-
     teristics of the Hydrogen Flame Detector," Varian Aerography Tech-
     nical Bulletin, pp. 133-167.

7.   Driscoll, J.N., et al., "Applications of a Gas Chromatograph Employ-
     ing an Integrated Photoionization Detector," American Laboratory,
     Jan 1980, pp. 84-93.

8.   Kern, H., M. Elser, "Trace Gas Analysis by a New High Sensitivity
     Thermal Conductivity Detector," Mikrochimica ACTA, Vol. I, 3-4,
     1978, pp. 319-328.

9.   Tsederberg, N.V., Thermal Conductivity of Gases and Liquids, MIT
     Press, Cambridge, Mass., 1965.

10.  Papoulis, A., Signal Analysis, McGraw-Hill, New York, 1977, pp. 324-
     327.
                                   67

-------
                                GLOSSARY
A/D converter - analog to digital converter - a device capable of con-
          verting an analog signal into an equivalent numerical value

algorithm -  a  numerical procedure used for solving a mathematical problem

anisotropic - an etch which has a directional dependence, usually caused
          by the crystalline nature of the substrate

CMOS - complementary metal oxide semiconductor - a semiconductor process
          used to produce devices with very low power dissipation

column - the long narrow tube in gas chromatography used to separate a
          gaseous sample

digital filtering - a procedure used to remove unwanted signals from a
          signal by numerical manipulation of the data

EPROM - electrically programmable read only memory - a program storage
          device

effective plates (N) - a measure of efficiency of a GC column

GC - gas chromatography - a process used to separate and measure the
          constituents of a gaseous sample

microcomputer - a small scale electronic programmable system for the
          manipulation of data

microprocessor - a single computational device which is the main arith-
          metic and logical element in a microcomputer system

                       -9
nl - nanoliter - 1 x 10   liter - a volume equal to a cube 100 ym on a
          side

OV-101 - a trademark for a silicone oil GC capillary column lining
          material

partition coefficient, K - the ratio of solute vapor to stationary phase
          concentration

partition ratio, k - the ratio of adjusted peak retention time to air
          peak time
                                    69

-------
plate height  (HETP) - a measure  of  efficiency of a column;  see Section
           4.2.3

program -  the  set of stored instructions for a computer  system

RAM - random access memory - the temporary data storage  elements in a
           computer system

SEM - scanning electron microscope

TWA - time weighted average
                                     70
                                         •6U.S. GOVERNMENT PRINTING OFFICE:1981--757-074/1 074

-------
DEPARTMENT OF HEALTH AND HUMAN SERVICES
    ;        PUBLIC HEALTH SERVICE
    .     CENTERS FOR DISEASE CONTROL.
N/  * .A.L INSTITUTE FOR OCCUPATIONAL SAFETY AND HEALTH
           ROBERT A. TAFT LABORATORIES
    4676 COLUMBIA PARKWAY, CINCINNATI, OHIO 45226


               OFFICIAL BUSINESS
           PENALTY FOR PRIVATE USE. $300
Third  Class  Mail
                                       POSTAGE AND FEES PAI
                                       US. DEPARTMENT OF HI
                                             HHS 396
                                                                     DHHS (NIOSH) Publication No. 81-115

-------