management that product qualit. should be the top priority, and that t.h failure of a company to produce qualit products is primarily a failure r management. It is mainly due to h efforts that there has been a majr shift in management attitudes towarr quality, and that numerous qualit improvement programs are now i force in North American industr: These programs have bee enthusiastically received by tt statistical community. In fact, or might say that this quality revolutio coming in response to the flood of hi, quality products from Japan, has dOl for the statistical community wh; Sputnik did for the aerospa( community. But where does the process contr engineer fit in? Even as recently; last year, polls conducted within son industrial process control grom placed product quality low on the Ii of reasons for implementing contr schemes (Schnelle and Richard 1986). Why are the ideas of proce control engineers so at odds with t present stress on quality? I feel that least some of the explanations for H attitude lie in the heavy emphasis 0 chemical engineering programs ha placed on petrochemical ope ratio where quality is somewhat less of concern than it is in special chemicals, electronics, biomateria etc. Another explanation might that, in the managerial structure many North American companil process control groups have become I isolated from the final customer, a rarely are able to relate the qual problems that these customers, experiencing back to the operation a control of the process. Finally, might also be claimed that chemi. engineers, with their inadequ; backgrounds in statistics, are i equipped to handle the noi, infrequent product quality data t.l typically is generat.ed off-line quality control laboratories.

Introduction

The term statistical process control has evolved to mean, in general terms, the use of statistical methods to improve process productivity and product quality. Included under the umbrella of SPC are all the statistical techniques involved with the design of experiments, the analysis of data, and on-line quality control methods, as well as the managerial aspects involved in effectively carrying out these improvement programs (Ishikawa, 1985). A great number of statistical design and analysis methods that are extremely useful in the chemical process industries have become available over the past fifty years. Examples include, factorial and fractional factorial designs (Box, Hunter and Hunter, 1978), optimal designs (Himmelblau, 1968), response surface methods (Davies, 1963; Hill and Hunter, 1966), evolutionary operation (EVOP) (Box and Draper, 1969), nonlinear and multiresponse estimation and design methods (Himmelblau, 1968), multivariate analysis (Anderson, 1958), and time series analysis (Box and Jenkins, 1970). An excellent application of

2. Some Examples The best way of introducing the on-line quality control problem is by way of a few examples. Consider the solution polybutadiene process described in Kelly et al. (1987). The process consists of a train of CSTR'S in which the butadiene monomer, a solvent, a chain transfer agent, and the components of the Ziegler-Natta catalyst system (a transition metal catalyst and a catalyst promoter) are fed to the first reactor in the train. The rubber is coagulated, and the solvent and unreacted monomer are recovered and recycled. The major disturbances in the process are due to impurity variations (ppm) in the feeds and buildup of impurities in the recycle. The process temperatures and flows are all controlled by conventional PID controllers, and the feedrates of the various inlet and recycle streams are controlled via a computer based material balance algorithm employing an on-line GC. However, the most important control problem in the process is that of controlling the final properties of the rubber (e.g., Mooney viscosity, etc.). These quality variables
recently, process control approaches based on discrete stochastic control theory have been used in a number of these situations (MacGregor and Tidwell, 1979; Kelly et aL, 1987). In the following sections we provide an overview of these latter two approaches, show where they overlap, and provide recommendations on their use.
3. Quality Control Charts 3.1 Shewhart Chart Most of the basic philosophy behind the use of quality control charts to monitor and control manufacturing processes were laid out by Shewhart (1931). The idea of simply plotting the data in some manner as soon as it became available, and observing trends and changes is very basic, and yet it is so often ignored. In order to help in assessing whether or not changes have occurred, Shew hart suggested plotting the data sequentially in time on a chart containing the target value and upper and lower action limits. This Shewhart chart (Figure 1) is probably still the most commonly used control chart. In the manufacturing industries samples of n units are usually taken periodically and both sample mean and the range of sample are plotted. The idea behind the test is that when the process is "in-control" the means should be independently and normally distributed about the target, and the variance should be constant. .:-_-,.,.UCL
_ _ _ _ _ _ _ _ _. target
small probability that they would be exceeded on chance alone if the process were "in-controL" Therefore, if the action limits are exceeded a change is called for to bring the process back to target (setpoint). This essentially constitutes a hypothesis test that the process mean is equal to the target against the alternative that it is not. Since this Shewhart procedure is not very sensitive to small deviations from target, it is common to augment it with runs tests, etc. (e.g., Ott, 1975). The purpose behind Shewhart's procedures is not simply to provide a decision mechanism for when to take feedback control action. Rather, by indicating when an "out-of-control" situation has occurred, it enables one to examine carefully the process data around that period of time in order to find an assignable cause. It is chiefly by this latter route that continual process improvements can be made (Ott, 1975; Ishikawa, 1976). 3.2 CUSUM Chart The cumulative sum tCUSUM) procedure was developed by Page (1954, 1961) and Barnard (1959) as a sequential Likelihood Ratio test for testing the hypothesis that the process mean is equal to the target value against the alternative hypothesis that it is not. Again it is assumed that the data are independently and normally distributed about a mean value p with a constant variance (021. In this procedure one plots the cummulative sum of the deviations from target since the last correction,

Sample Number

Figure 2:

Plot with V-mask

3.3 EWMA Chart The exponentially weighted moving average IEWMAJ control chart was proposed by Roberts (1959). A more recent discussion of this procedure is given by Hunter (1986). In this procedure the EWMA (Y I ) of the observations is plotted, where

YI "" (l - OJ

""OY 1- z+(1-6lY I
the EWMA parameter (0 < e < 1) which determines how fast one discounts past data. Whenever the EWMA exceeds some upper or lower control limit an adjustment is called for.
Note that as e tends to zero only the current point is weighted and the EWMA chart will be equivalent to a Shewhart chart, and as 0 tends to unity the EWMA approaches a cumulative sum. Although in practice the value of e is often selected from experience (8 "" 0.8 being a common choice) its optimal value can be estimated from the data themselves. The original justification of Roberts for the EWMA chart was rather intuitive.
Figure 3: Simple Process Discrete Process Models
Linear difference equation and discrete (pulse) transfer function mode Is are we II known to process control engineers, and so we shall only review them briefly here. Furthermore, in the subsequent
sections we are only going to be concerned with the simplest forms of these modeIs.
4.1 A Pure Gain (or Steady-State) Process

1 - oz- 1

output (Y) follows such a process during "in-control" periods. However, in the process industries such an independent behaviour from time to time is not typical of most process disturbances. Disturbances entering the process are often persistent in nature, such as variations in raw materials. The properties of these materials tend to drift high or low for many time periods. Furthermore, most processes are continuous in nature and disturbances entering at v~rious points will pass through part of the dynamics of the process, and continue to affect the output for several time periods. Therefore, the disturbance as it appears in the output measurement (D t in Figure 3) will, in general, not be just random white noise, but will exhibit a dependence upon past values, that is, it will be autocorrelated. Discrete time series models capable of represen ting such au tocorre lated behaviour were first introduced by Yule (1927) and more recently treated thoroughly by Box and Jenkins (1970l. The basic idea is that by starting with a white noise sequence and passing it through a digital shaping filter a highly auto'correlated disturbance process can result at the output. Consider, for example, the first order

5.0r----~~---~----~----.,
2O'0r--~-~---r---,----.--..,---.----,

3.0 2.0

15.0 10.0

0.0 -1.0

-2.0 -3.0
-15.0'---~-~--~-~--~-~--~--'
-5.0'-----~----~----~:__--_=_~

Figure4:

Stationary Autoregressive Disturbance (<I> = 0.8)

TIME Figure 5:

Random Walk Disturbance.

D'+lI'

- 8)[D,

+ 8D'_1

and predictions have proved so successfuL 5.3 Randomly Occurring Deterministic Disturbances This same class of ARIMA models (19) can also be used to model deterministic disturbances which occur randomly but infrequently in time such as steps, ramps, or exponential changes (MacGregor, et aL, 1984). The difference in the models lies in the probability distribution of the random process {a,; t = 1,2,. }. For deterministic disturbances these at's are zero most of the time, except at the occurrence of a change. Since the. minimum mean squared error predictors for these disturbances are independent of the nature of the probability distribution of the at's (only requiring that it be symmetric) then the prediction equations are identical for the same model structure. The implication is that there is no difference between the design of optimal controllers for stochastic or deterministic disturbances. The model for randomly occurring step changes is given by
by a first order moving average (cf equations (16.

+ 8 D'_2 + 8 D'_3 +. J

A general class of stochastic disturbance model is given by the integrated ARMA model (19)

Minimum Variance Control

Given that a combined process dynamic and disturbance model of the system has been identified, that is

( z -1) z -b

8( -1)

8(z - 1)

<1>( z- 1) V d
The polynomial orders (p,d,q) and the parameters of a model from this class which characterizes the disturbance in any given process can usually be identified directly from data collected from that process (Box and Jenkins, 1970). This area of identification is a mature field in hoth the statistics and the control literature. For nonstationary disturbances the degree of differencing (d) is usually not greater than one. An important result for nonstationary ARIMA processes in this class (d = 1) is that, as one samples them less and less frequently (i.e., the sampling interval T gets large), they all tend to the limiting first order integrated moving-average process in equation (15) (MacGregor, 1976). Since most SPC situations involve the use of infrequently measured laboratory data, it is not surprising that this disturbance process occurs so frequently in practice, and that EWMA control charts

one can easily design a control algorithm to satisfy a desired objective. In the case of product quality control a very reasonable objecti ve is to try to minimize the variance of the output deviations from the target or setpoint. Therefore, in the following sections we examine the structure of minimum variance controllers in several special situations that arise commonly in SPC problems. 6.1 No Process Dynamics Consider the situation where the process attains a steady-state in the interval between sampling instances, that is. (22) As mentioned previously, this steadystate or pure gain model is the rule in most parts manufacturing applications, and may be reasonable in
the same structure as that for a random walk. If white measurement noise is present at each interval, then the process is again well approximated
8.0 r--~----'--~--'---~--,.--r---,

-6.0'-_~_~

_ _.J.L_~_ _~ _ ~_ _~_.J 150.0 200.0 250.0 300.0 350.0 400.0
TIME Figure 6: Integrated Moving-Average Distrubance
the process industries when the sampling interval is long re lati ve to the process time constants. For perfect regulatory control, in the face of disturbances (D,), we should choose u, to set the deviation from setpoint, Y'+l, equal to zero, that is, we should set

, - - g

often occurring in the parts manufacturing industries where there is a distinct non-zero cost associated with taking a control action (e.g., stopping a machine and readjusting it), and (ii) another more often occurring in the process industries where there is no direct cost associated with taking a control action (e.g., resetting the setpoint of a P1D controller).

6.1.1 Non-Zero Costs:

et al. (1963, 1974) using dynamic programming methods. By considering that the loss associated with being off-target (Y, '" 0) is equal to cY,2, and the non-zero cost of making an adjustment is C, Box et al. also derived expressions for the placement of the upper and lower control limits. The placement of these limits depends upon relative costs (c,C) and the disturbance characteristics (8). In summary, it has been shown that in the situation where there are both no process dynamics and non-zero control costs some of the traditional control chart methods indeed are optimal policies.

6.1.2 Zero Costs:

However, this control action is not realizable because it involves a future unknown value of the disturbance. Therefore, the best we can do is to minimize the variance of the output deviations from setpoint. It is readily shown (Box and Jenkins, 1970) that this is achieved by replacing D,+ 1 by its minimum variance prediction (Dt+ 1ft) i.e.,

1 u, - - - D g

It is obvious that the control action depends largely upon the nature of the disturbance D,. we could consider many m';f,llro'in(,e models at this point, we concentrate on the first order int:e"ralted moving-average model (15) it arises so frequently in SPC Recall that the minimum VaI'la'"Ce predictor for this disturbance is the exponentially weighted IEWMAI predictor in equations (16,17,18). Hence minimum variance (MY) controller given by
In this situation the optimal control policy (on the basis of maximizing profit) will not be to implement the MY control action (26) at every sampling interval. For small changes, the cost of stopping the production to make the changes will usually exceed the increased profit resulting from tighter control. The decision on whether or not to take action must obviously depend upon the relative cost associated with being off-target versus that associated with making a change. If control action is taken only infrequently, then between actions the process is running open-loop and thus (27) Y =D

1D =- - [8) , +8D g

+ 8 D 1- 2+. J

(1_8z- 1)

-'----D

(l - 8)

At some point, where the predicted variation from target becomes sufficiently large a control action (26) must be taken. The resulting optimal control strategy therefore will be equivalent to an EWMA chart in which the EWMA predictions of the output deviations from target (Y,+ 1ft =D,+1ft) are plotted. When the EWMA exceeds some control limit an action is taken, and the EWMA is restarted. Note that when the parameter 8 of the disturbance model is equal to zero then Yt+ 1ft = Y" and the optimal control strategy is equivalent to that using a simple Shewhart control chart. Although the optimal "non-zero cost" control strategy has been justified above on an intuitive basis, it has in fact been rigorously developed by Box
Now consider a second situation, more common to the process industries, in which there is no cost associated with taking a control action. In this situation, there is no justification in waiting for a sufficiently large deviation from target before taking control action. Rather, to minimize the variance of the output deviations, the action (26) should be taken at. every sampling interval. However, in this situation, the disturbance D" will no longer be simply equal to the measured output as in equation (27). Rather, the output will be a function of both the past control action and the disturbance, and hence the disturbance must be inferred by subtracting off the effect of past control actions, Le.,

pinpoint the cause of the disturbances and perhaps eliminate or minimize such disturbances in the future. Of course, this is where real process improvement is made. Very efficient process control schem~s often serve as bandaids that hide things that should be improved at the process level. To correct this control charts and other SPC methods could be used more frequently for analyzing control system performance, and as diagnostic tools. 8. Conclusions In the introduction to this paper, it was stressed that there are many aspects to SPC. I have concentrated on only one of these in this paper; namely on the use of SPC charts for on-line control of product quality. Since statisticians and control engineers appear to have very little overlap in their knowledge base, this paper has used simple stochastic control theory to try to help bridge this gap. It has been shown that control chart schemes are indeed optimal control schemes in many SPC situations, particularly in the parts manufacturing industries. However, to blindly extrapolate these methods to the process industries where process dynamics and control costs are very different, may not be efficient. On the other hand, there is much that the process control engineer can learn from the SPC movement. At the very least, he should capitalize on management's current commitment to improving product quality and productivity. References
Anderson, T.W. (958), An Introduction to Multivariate Statistical Analysis, Wiley, N. Y.
Astrom, K.J. and Wittenmark, B., 0984}, Computer Controlled Systems, Prentice-Hall, N.J.
Bather, J.A. (1963), Control Charts and the
Minimization of Costs, J. Roy. Statis. Soc. B, 25

49-80.

Barnard, G.A. (1959), Control Charts and Stochastic Processes, J. Roy. Statis. Soc., B21, 239-271.
Box, G.E.P. and Jenkins, G.M. (1963), Further Contributions to Adaptive Quality Control: Simultaneous Estimation of Dynamics: Nonzero Costs, Bull. Int. Statist. Int., 34th Session, 943-978. Box, G.E.P., Jenkins, G.M. and MacGregor, J.F.
(1974), Some Advances in Forecasting and Control, Part II, J, Roy. Statist. Co, C (Applied

Statistics), 23,158-179.

Box, G.E.P., and Jenkins, G,M. (1970), Time Series Analysis: Forecasting and Control,

Holden Day, S.F.

Box, G,E.P. and Draper, N.R. (1969), Evolutionary Operation. A Statistical Method for Process Improvement,Wiley, N.Y_ Box, G.E.P., Hunter, W,G. and Hunter, J.S, (1978), Statistics for Experimenters, Wiley, N.Y. Cuttler, C.R. and Ramaker, B.L. <l976}, Dynamic Matrix Control, 86th AIChE Meeting, April,1976. Deming, W.E. (1967), What Happened in Japan, Industrial Quality Control, 24, 89-9'3. Deming, W.E. {l972}, Report to Management, Quality Progress, 5, Deming, W.E. (1975), On Some Statistical Aids

Towards Economic Production, Interfaces, 5, 1-
Deming, W.E. (982), Quality, Productivity and Competitive Position, MIT CAES.
Deming, W.E. {1986), Out of Crisis, MIT CAES.
Harris, T.J. and MacGregor, J.F. (1987), Design
of Multivariable LQ Controllers Using Transfer Functions, Amer. Inst. Chem. Eng., J., 33, Oct.
issue. Hill, W,J. and Hunter, W.G. (966), A Review of Response Surface Methodology, Technometrics, 8,571.
Hirnrnelblau, D.M. (1968), Process Analysis by Statistical Methods, Wiley, N.Y. Hunter, J.8. (1986), The Exponentially Weighted Mouing-Average,J. Quality Tech., 18,203-210. Ishikawa, K. (1976), Guide to Quality Control, Asian Productivity Organization, Tokyo.
Shewhart, W.A. (1931), Economic Control of Quality, Van Nostrand. Taguchi, G. and Wu, Y. (1985), Intoduction to Off-Line Quality Control, Central Japan Qualtiy Control Association. Van Dobben de Bruyn (1968), Cummulative Sum Tests, Griffin, London.
Ishikawa, K. (1985), What is Total Quality Control? (The Japanese Way). Prentice-Hall, N.Y.
Kelly, 8.J., MacGregor, J.F. and Hoffman, T.W. (1987), Control of a Continuous Polybutadiene Polymerization Reactor Train, Can. J. Chern. Eng.,.65, Aug. issue. Lucas, J.M.,.(1976), The Design and Use of VMask Control Schemes, J. Quality Tech., 8, 1-8. Lucas, J.M. and Crosier, R.B. (982), Fast Initial Response for CUSUM Quality Control Schemes, Technometrics, 24,199-205. MacGregor, J.F. (1976), Optimal Choice of Sampling Interval for Process Control, Technometrics
Woodall, W.H. (1986), The Design of CUSUM Quality Control Clw.rts, J. Qual. Tech., 18, 99102.
Review of Mathwriter, an Equation Writing Software Package for the MacIntosh
by Bruce A. Finlayson There are very few computer programs that revolutionize how I work, but this is one of them. I have used the program for about nine months and have found it easy to use; it was designed with a technical writer in mind. The equations appear on the screen as they will in the document; this makes for easy editing. The program can, of course, do all the standard symbols: integrals, summations, fractions, and matrices. As an illustration of the excellent design, when an integral is chosen the computer immediately moves to the location for the lower limit of integration and automatically changes the font size to a smaller size. Superscripts and subscripts can be chosen with a mouse, and the font size is automatically reduced from what it was, as you would like. Greek letters are displayed at the bottom, and can be chosen with a mouse. This is easier than changing the font back and forth between Greek and Roman letters. There are many options available. For example, pallets containing all sorts of mathematical symbols can be added to the bottom as well. Equations, once formed, can be edited, copied, and otherwise changed. They can also be pasted into MacWrite and Word.

they include only enough information as needed to recognize the equation. Algebraic rearrangement can be done easily; the result is text that is more clear, with simpler steps, and is easier to follow. I have used the program extensively while writing a book, and the program is a virtual necessity now. The software is available for $49.95 from Cooke Publications, P. O. Box 4448, Ithaca, New York 14852. "Heterogenous Azeotropic Distillation~ Homotopy-continuation Methods"
J.W. Kovach, III, W.O. Seider
"Vse of 2D-Adaptive Mesh in Simulation of Combustion Front Phenom'ena"
J. Degreve,P. Dimitriou, J. Puszynski. V. Hlavacek,S. Valone, R Behrens
"The Dominant Time Constant for Distillation Columns"

S. Skogestad, M. Morad

"Contribution of Multiple Scattering to Light Transmission by a Collimated

Beam"

S.S.Ou,J.D.Seader
"An Algorithmic Procedure for the Synthesis of Distillation Sequences with Bypass" R. Wehe, A.W. Westerberg "Optimal Reflux Rate Policy Determination for Multicomponent Batch Distillation Columns"
U.M. Diwekar, R.K. Malik, K.P. Madhavan
"Inversion of Sparse Matrices by a Method based on Graph Theory"
G. Samuel, M. Pollatschek, E. Kehat
Mrs. Elizabeth Hughes to Receive the Memorial Issue of Computers and Chemical Engineering in Honor of Prof. Richard R. Hughes at the CAST Division Dinner, November 18,1987 The Memorial Issue, edited by Prof. Warren D. Seider, will be presented to Betty Hughes at the CAST Dinner in the New York Hilton on November 18. It contains articles by many close associates of Dick Hughes and is summarized below: In Memory of Richard R. Hughes
Copies are available from Pergamon Press for $25.00 per copy.
"Synthesis and Sizing of Batchl Semicontinuos Processes: Single Product Plants" N.C.C. Yeh, G.V. Reklaitis "DESIGN-KIT: An Objective-oriented Environment for Process Engineering"
G. Stephanopoulos, J. Johnston,T. Kriticos,
The Intel ipSC/2: The First Concurrent Supercomputer for Production Applications On August 31, 1987, Intel announced what it considers to be the next generation of concurrent computers, the ipSC/2. The Intel ipSC/2 offers full 32-bit node a.rchitecture, up to a gigabyte of memory, concurrent development tools (Concurrent Workbench ITMll, and new communications between nodes' (Direct-Connect Routing). A standard system consists of 32 to 128 nodes. At each node, there is a Direct-Connect ITMl Routing Module; 1, 4, 8, or 16 megabytes of modular memory; an 80386 processor; an 80387 floatingpoint processor; and a UNIX-based development environment. An extended memory ipSCI2 MX system has 16 megabytes of memory per node, expandable to 64 nodes. A vector ipSCI2 VX system consists of 16

"Heuristic Synthesis of Sloppy Multicomponent Separation Sequences," by Cheng and Liu "Synthesis and Optimal Design of Alternative Sequences for Separating Heterogeneous Azeotropic Mixtures," by Ryan and Doherty "Recent Advances in the Analysis of Heat Recovery Problems," by Jones and Rippin "Synthesis of Utility Systems Integrated with Chemical Processes," by Colmenares and Seider "Process Integration Subject to Match Constraints," by O'Young and Linnhoff "Design and Analysis of Heat Integrated Distillation Sequences for Multiperiod Operation," by Paules and F'loudas
4. Artificial Intelligence in Process Engineering. H. Dennis Spriggs (Chairman), Linnhoff March, P. O. Box 2306, Leesburg, VA 22075, (703) and V. Venkatasubramanian (Vice Chairman), Department of Chemical Engineering, Columbia University, New York, NY 10027, (212) 280-4453.
"Heat Exchanger Network Synthesis: A Knowledge Engineering Approach," by Fan "An Expert System for Designing Distillation Plates," by Davis, Myers, and Herman

"STES:

A Separation Process Expert System," by Netterfield and Sunol

"RIP:

A Prototype Expert System for Retrofitting Chemical Plants," by Nelson and Douglas
New York City AIChE Meeting
(November 15-20, 1987) Area lOa Sessions 1-2. Design and Analysis I and II. Richard S. H Mah (Co3. Computer Aided Design of Batch Processes. Kris R. Kaushik (Chairman), Shell Oil Company, P. O. Box 2099, Houston, TX 77252-2099, (713) 241-2098 and Malcolm L. Preston (Vice Chairman), Imperial
"Design of Polymer Composites: A Blackboard Approach," by Vellkatasubramian, Lee, and Gryte
"POPS: The Prototype. Operating Procedure Synthesis Program," by Fusillo
Joint Areas lOa and lOb Session 1. Integration of Process Design and Control. Bradley R. Holt, Department of Chemical Engineering, BF-I0, University of Washington, Seattle, WA 98195, (206) 543-0554 and W. David Smith (Vice Chairman), Polymer Products Division, E. 1. DuPont de Nemours and Co., Wilmington, DE 19898, (302) 7721476.

"On-Line Optimization of Complex Process Units: A Comparison of Centralized versus Distributed Approaches," by Darby and White "Global Optimization of Nonconvex MINLP Proqlems in Process Synthesis:' by Kocis and Grossman "Simultaneous Heat Integration and Optimization of Distillation Sequences," by Lin and Prokopakis "An MINLP Formulation for the Synthesis of Continuous Pressure Heat Integrated Distillation Sequences," by Floudas and Paules "Process Optimization Through Symbolic Computation," by Sunol
"Optimization Model for Long Range Planning In the Chemical Industry," by Grossmann, F'ornari, and Chatrathi "Refinery Planning, Scheduling, Monitoring, and Control: Quantitative Capability for Management Requirements:' by Dorweiler and Bryant "Productivity Analysis of a Large Multiproduct Batch Processing Facility," by White "Planning and Scheduling ofBateh Operations," by Thomas and Shobrys "Refinery Performance and Refining Methods," by Dorweiler and Bryant
further details concerning Area lOb sessions and scheduling, please contact Yaman Arkun (Chairman, Area lOb), Department of Chemical Engineering, Georgia Tech, Atlanta, Georgia 30332, (404) 894-2871.
"Approximate Method for Scheduling Multiproduct Multiline Operations," by Ford
5. On-Line Fault Administration. Mark Kramer (Chairman), Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, (617) 253-6508 and J. F. Davis (Vice Chairman), Department of Chemical Engineering, Ohio State University, 140 West 19th Avenue, Columbus, Ohio 43210.
"Operator-Assisted Learning in Expert Systems for Process Fault Diagnosis," by Venkatasubramanian "A Connectionist Expert System Approach to Fault Diagnosis in the Presence of Noise and Redundancy," by Gallant "Expert System in a Wastewater Treatment Process Diagnosis," by Marcos "An Operator Aid for Analysis of Disturbances in Distillation Columns," by Andow

"MOLDOCTOR:

494-2257; and Neal R. Amundson (Honorary Chairman), Department of Chemical Engineering, University of Houston, Houston, TX 77004.
"Models: Good and Bad," by Aris "Chemical Reaction Engineering in Practice," by Krambeck "Computer-Aided Mathematical Analysis," by Scriven "Modern Analysis of Steady-State Multiplicity," by Luss "The Role of Applied Mathematics in Polymerization Engineering," by Ray "The Impact of Applied Mathematics in Computer-Aided Design," by Reklaitis "Some Peculiarities in the Relationship between Mathematics and Chemical Reaction Engineering," by Feinberg "Reflections," by Amundson

Six technical sessions are planned, each conducted by a ChairmanRapporteur, containing presentations of five to six papers of 30 minutes duration, including discussion. Following the successful poster session at PSE '85, a similar session is planned this time. The Conference proceedings will be published. An exhibition relevant to the themes of the Conference will run concurrently. The main conference themes are: Process Control and Optimization. Benefits Assessment. Operator/Process Interface. Plant-wide Systems Artificial Intelligence. On-Line Expert Systems. Design/Synthesis Applications Batch Process Design and Operation. Including Operability Considerations. Scheduling Applications. Batch Process Control Industrial Applications. Case Studies with Benefits Through Applications ofpSE Failure Analysis in Design. Reliability/Availability Theory for Process Systems. Applications to Process Design " Hazard Identification Techniques Design ofFlowsheets. Retrofitting '" Synthesis " Operability. Minerals, Solids and Other NonPetrochemical Processes Modelling New Models and Algorithms II Process Identification
Education in PSE., Undergraduate/Postgraduate '" Continuing Education
The timetable for authors is: August 31,1987 - Abstract to address overleaf December 31, 1987 - Full paper for refereeing April 30, 1988 - Final manuscript The organizing committee is:
Dr. M. L. Brisk, Chairman
Name: Title: Affiliation: Postal Address: I would like to submit a paper (abstract attached)
of Yladison, Madison, WI 53706, (608) 262-3641. Design of Integrated 5. Biotechnology Process Systems. George Stephanopoulos (Chairman), Department of Chemical Engineering 66-562, Massachusetts Institute of Technology, Cambridge, YlA 02139, (617) 253-3004 6. Design of Polymer Process Systems. Michael F. Malone (Chairman), Department of Chemical Engineering, University of Massachusetts, Amherst, MA 01003, (413) 545-4869 and Kendree J. Sampson (Vice Chairman), Department of Chemical Engineering, Ohio University, Athens, OH 45701, (614) 593-1503, For further details concerning Area lOa sessions and scheduling, please contact Michael F. Doherty (Area lOa Chairman-Elect), Department of Chemical Engineering, University of Massachusetts, Amherst, MA 0 1003, (413) 545-2359. Area lOb Sessions 1-2. New Developments in Process Control I and II. John W. Hamer (Co-Chairman), Research Laboratories, Eastman Kodak Company, B82 1st Flood, Rochester, NY 14650, (716) 477-3740 and Professor W. Harmon Ray (CoChairman), Department of Chemical Engineering, University of Wisconsin,' 1415 Johnson Drive, Madison, wt 53706, (608) 263-4732. 3. Robustness and Modeling Issues in Process Control. Professor Ahmet N. Palazoglu (Co-Chairman), Department of Chemical Engineering, University of California, Davis, CA 95616, (916) 752-8774 and Professor ,Jeffrey C. Kantor (Co-Chairmanl, Department of Chemical Engineering, University of Notre Dame, Notre Dame, IN 46556, (219) 239-5797.

Australia Pty Ltd, Joint

(Acceptance of papers is conditional upon at least one author attending the Conference to present it.) I am considering attending and would like to recei ve the Second Announcement.

PSE '88

Professor J. D. Perkins, University of Sydney, Joint Chairman
Mr. J. E. Atkins, CSR Ltd.
Dr. G. W. Barton, University of Sydney Dr. I. Cameron, University of Queensland
Dr. R. D. Johnston, University of New South Wales
Mr. G. D. Kelly. BHP Steel International Group
I may be interested in post-conference tours in Australia to: The Great Barrier Reef Queensland's Gold Coast Central Australia Please mail to:
PSE '88 Conference The Institution of Engineers, Australia 11 National Circuit Barton, ACT 2600 Australia
Dr. D. Sutherland, csmO Division of Mineral Engineering
Professor D. Depeyre, France Dr. W. B. Earl, New Zealand

Yes Yes Yes

Professor G. V. Reklaitis, United States of America Professor R. W. H. Sargent, United Kingdom ProfessorT. Takamatsu,Japan Dr.J. D. Wright ,Canada
Thinking of Attending? If you are considering attending PSE '88, whether or not you plan to submit a paper, please return the Registration of Interest slip (duplicated below) now to ensure that you receive a copy of the Second Announcement and detailed Program.
Washington, D.C., AIChE Meeting
(November 27-December 2, 1988) Area lOa Sessions 1-2. Process Synthesis I and II. James M. Douglas (Chairman), Department of Chemical Engineering, Cniversity of Ylassachusetts, Amherst, YlA 01003, (413) 545-2252. 3-4. Design and Analysis I and II. G. V. Reklaitis (Chairman), School of Chemical Engineering, Purdue University, West Lafayette, IN 47907, (317) 494-4089 and Professor Ross E. Swaney (Vice Chairman), Department of Chemical Engineering, University

Registration ofInterest

Please make a Xerox copy and complete the details of the following. Send it to the address shown.
4. Unsolved Problems in Process Modeling, Optimization, Control and Operations. Professor Christos Georgakis (Chairman). Chemical Process Modeling and Control Research Center, Lehigh University, Bethlehem, PA 18015, (215) 758-4781 and Dr. Jorge Mandler (Co-Chairman), Air Products and Chemicals, P.O. Box 538, Allentown, PA 18015, (215) 4813413. 5. Adaptive Control. Professor B. E. Ydstie (Chairman), Department of Chemical Engineering, University of Massachusetts, Amherst, MA 01003, (413) 545-2388 and Professor C. Brosilow (Co-Chairman), Department of Chemical Engineering, Case Western Reserve University, Cleveland, OH 44106, (216) 368-4180. 6. Expert Systems in Process Control. Professor Bradley R. Holt (Chairman), Department of Chemical Engineering BF-IO, University of Washington, Seattle, WA 98195, (206) 543-0554 and Dr. Carlos Garcia (CoChairman), Shell Development Company, Westhollow Research Center, Houston, TX 77001, (713) 4938873. Joint Session Between Areas lOb and 15c 7. Control of Biochemical Systems. Professor Karen McDonald (Chairman), Department of Chemical Engineering, University of California, Davis, CA 95616, (916) 752-0400 and Prof. Anil Menawat (Co-Chairman), Department of Chemical Engineering, Tulane University, New Orleans, LA 70118, (504) 865-5772. For further details concerning Area lOb sessions and scheduling, please contact Yaman Arkun (Chairman, Area lOb), Department of Chemical Engineering, Georgia Tech, Atlanta, Georgia 30332, (404) 894-2871.

2.2. Uncooled AlGaInAs-lasers for direct 10 Gbit/s modulation Fibre based optical metropolitan and access networks require low cost 1.3 m transceivers comprising transmit, receive and electrical multiplexing functions. These transceivers have to comply to small-form-factor module standards, should have an electrical power consumption as low as possible, and must meet the operation specifications within the full temperature range from 0 to 85 C without any active cooling. The lasers to be implemented in such modules must therefore show low threshold currents, high temperature stability, and depending on the application 2.5.10 Gbit/s modulation capability at operating currents as low as possible. For 2.5 Gbit/s applications such as e.g. the LX4-standard GaInAsP/GaInAsP-BH-laser diodes fulfil these demands. At higher bitrates the relatively low characteristic temperature of GaInAsP/GaInAsP-lasers impedes high modulation bandwidth at temperatures higher than about 70 C. Here lasers using AlGaInAs/AlGaInAs-MQW-structures as active layers are known to be advantageous due to their higher conduction band (lower valence band) discontinuity. However all Al-containing layers are very sensitive to oxidization. Hence in order to avoid any detrimental oxidization effects it is essential that the active layers do not get into contact with ambient air during the processing of such lasers. This means that standard BH-lasers with pn- or semi-insulating blocking layers cannot be fabricated using AlGaInAs/AlGaInAs-active layers. Although some new approaches to realize BH-lasers with Al-containing active layers have been published recently there is still a long way to go until these lasers can be commercialized. Much easier is the fabrication of ridge-waveguide lasers with Al-containing active layers, because these laser structures do not require to etch the active layer. So oxidization effects can be excluded here. Therefore to our knowledge all commercially available uncooled 1310 nm-AlGaInAs/AlGaInAs-lasers are based on a ridgewaveguide structure. AlGaInAs/AlGaInAs-RW-lasers with surface n-contacts have been fabricated recently, as surface n-contacts allow easy on-chip direct RF-measurements and are furthermore a pre-condition for flip-chip-mounting of the devices. Being much smaller in size as the standard backside contact the surface n-contacts have to be optimized towards a very small specific contact resistance in order to achieve the same laser series resistance, and the results of such an

L400pm

2OC 5OC 7OC

________________

backside n-contact surface-n-contact

0 > 0.5

current (mA]
Fig. 1: VI-curve of 1310 nm AlGaInAs/AlGaInAsRW-FP-lasers with 0.3/0.9 facet reflectivities.
Fig. 2: Optical output power of AlGaInAs/AlGaInAsRW-FP-laser.
optimization is shown in Fig. 1, where identical PI-curves could be obtained independent of the n-contact type used. 250 m long devices with 0.3/0.9 facet reflectivities show high output power and a characteristic temperature >80 K. With 200 m long 0.6/0.9 facet reflectivity devices an even higher value of about 90 K could be obtained. Compared with a typical value of around 50.60 K for standard 1300 nm GaInAsP/GaInAsP-laser devices this clearly demonstrates the superiority of the AlGaInAs/AlGaInAs-material system for high temperature operation.
Proc. of SPIE Vol. 5956 59560I-2

3OmA 4OmA

2OmA 25mA 3OmA

4OmA 5OmA

c -20 0

-30 -40

frequency (GHz]
Fig. 3: Frequency response of AlGaInAs/AlGaInAs-RW-FP-laser with 0.3/0.9 facet reflectivity at 20 oC (left) and 90 oC (right) (currents include the threshold current).
The corresponding small signal response behavior of the 250 m long devices is shown in Fig. 3. The modulation bandwidth at 80 mA operation current amounts to >14 GHz at 20 C and >9 GHz at 90 C, which shows the capability for uncooled 10 Gbit/s operation. The record value for 1300 nm AlGaInAs/AlGaInAs-RW-FP-lasers has been published recently by Paoletti et al. [6]. 200 m long HR/HR-coated devices with standard backside n-contact show a modulation bandwidth of 14 GHz at 85 C at an operation current of 80 mA. The characteristic temperature of the device amounts to about 95 K. One remaining challenge is the provision of a suitable coating of the facets, which prevents their oxidation and correlated device degradation.
2.3. Tapered Fabry-Perot lasers Large alignment tolerances (>1 m) are a prerequisite for low-cost packaging of lasers, and such tolerances can be achieved by the monolithic integration of a spot-size converter ('taper') with the laser. One particularly attractive concept is an all-active taper [7-11], i.e. a tapered Buried-Heterostructure (BH) laser, which is on the one hand very short (and thus offers high yield per wafer) and which can be fabricated essentially without additional processing steps (i.e. at the cost of a standard BH-laser). Due to the fact that the taper is part of the FP-laser cavity there is a trade-off between a small optical far field on the one hand and a low threshold current in combination with good high temperature behavior on the other hand. Thus it is necessary to find a design, which leads to a good balance between acceptable far field and desirable threshold current. A schematic structure of a corresponding tapered GaInAsP-BH-FP-laser is shown in Fig. 4. The active layer consists of compressively strained GaInAsP quantum wells separated by GaInAsP barriers with tensile strain and embedded in a matching GaInAsP waveguide. Fabrication of the laser structure requires three MOCVD growth steps: The first growth run provides the laser heterostructure, and a subsequent reactive ion etching process creates the tapered laser stripes. These are then selectively overgrown by InP pn blocking layers, and the final growth step provides the p-InP cladding and the p-GaInAs contact layers.

Fig. 4: Schematic view of an all-active tapered GaInAsP-FP-BH-laser.
Proc. of SPIE Vol. 5956 59560I-3
The optical power emitted from the front facet of a 400 m long as-cleaved tapered 1300 nm GaInAsP-BH-FP-laser is shown in Fig. 5 (left), which illustrates low threshold currents (about 6 mA at 20 C and 25 mA at 90 C, respectively) and high output power without facet coating. The performance of the taper is illustrated by the far field shown in Fig. 5 as well (right). It is essentially circular with a FWHM-value of 16 only, which enables low-loss coupling into optical fibres or optical board waveguides.

205070 90CC

L=400pm

lateral

angle (0]
Fig. 5: Typical optical output power and far field of tapered 1300 nm GaInAsP-BH-FP-lasers.
2.4. Tapered DFB lasers The concept illustrated for the case of all-active tapered BH-FP-lasers can also be transferred to DFB-lasers. However, the effective refractive index of a tapered laser (or waveguide) structure exhibits a spatial variation, and as a consequence the grating period of a corresponding laser requires a counteracting (or compensating) spatial variation in such a way that a uniform Bragg wavelength is obtained all along the tapered stripe [11]. Such 'chirped

1=1 OOmA

FWHM 14

-40 -60 1525

0.4 0.2

'. 15 0

caIFWHM:2O

-40 -20

wavelength [nm]
Fig. 6: Typical optical output power, emission spectra and far field of tapered 1550 nm GaInAsP-BH-DFB-lasers.
gratings' can be fabricated using a direct electron beam writing process followed by reactive ion etching. The characteristics (output power, side mode suppression ratio, far field) of all-active tapered 1.55 m-DFB-BH-lasers with AR-coated facet are illustrated in Fig. 6, and the very good performance is obvious. As the fabrication of chirped gratings requires time-consuming electron beam writing this approach is expensive. A cost-efficient alternative is the use of a bent active tapered structure in combination with a uniform DFB grating, and corresponding devices have shown far field angles comparable to those shown in Fig. 5, very high single-mode yield even in the absence of any antireflection coating of the facets, and no degradation of the spectral characteristics with increasing operation temperature. Further details will be published elsewhere [12].
Proc. of SPIE Vol. 5956 59560I-4
Due to the already mentioned threshold current / far field trade-off the lowest obtainable optical far field angles of allactive tapered DFB-lasers are restricted to values of about FWHM > 15. For applications, where even smaller optical far fields are required, only DFB-lasers with a monolithically integrated additional passive optical taper can be used. In this type of DFB-laser the performance of the DFB-laser and that of the taper can be optimized largely independently of each other. Thus very small far field angles are possible, however, the fabrication of these lasers is much more complicated and as a consequence considerably more expensive.

2.5. Complex-coupled DFB lasers Another approach towards reducing system cost is the use of complex-coupled (CC) DFB lasers instead of standard index-coupled (IC) ones. IC-DFB-lasers are rather sensitive to optical feedback and corresponding transmitter modules normally require an optical isolator, which raises their cost considerably. On the other hand, CC-DFB-lasers with high coupling strength L of the gratings are significantly more robust against optical feedback [13] so that optical isolators can be omitted in many cases. This is explained in more detail in the following section, where CCand IC-DFB lasers with otherwise identical structure are compared with respect to their device characteristics, and in a 2.5 Gbit/s transmission experiment as well. Fabrication details of the lasers have been reported elsewhere [14] and will not be repeated here. Both laser types exhibit >40 dB side mode suppression ratio, and the threshold currents (at 20 C) are 6 mA and 15 mA for the IC- and CC-lasers, respectively. The increased threshold current (and lower output power at identical operation current) in the CC-DFB lasers is attributed to the smaller active volume and higher internal losses. The coupling strength of the ICDFB laser amounts to i = 50 cm-1, resulting in a characteristic value of L = 2, (significantly) higher L-values are not acceptable due to the onset of spectral hole burning [15]. For the CC-DFB laser the real and the imaginary part of the coupling parameter are estimated to be i = 215 cm-1 and g = 27 cm-1, respectively ( L = 9+1i). The lasers have been packaged into modules with a constant coupling loss (4 dB) between laser and a tapered optical fibre. Via an optical coupler the lasers have been connected to a setup providing controllable feedback strength (Fig. 7). The
Fig. 7: Measurement setup for the characterisation of feedback sensitivity.
feedback delay corresponds to 5 m optical fibre, and the feedback strength (reflected power R) is measured in the fibre, i.e. R is not affected by the coupling losses at the fibre/laser interface. The effect of feedback on the spectral linewidth and RIN under dc conditions is shown color coded in Fig. 8 for various feedback strengths (given on the y-axis). As already described in the literature [16], three distinct regimes can be observed for IC-DFB lasers: For low feedback power (32 dB) the laser exhibits single mode behaviour (sharp line at 1554.4 nm) with a small linewidth and low RIN values below 145 dB/Hz. For moderate feedback strength between 24 dB and 12 dB the spectrum of the IC-DFB laser starts to split into harmonic spectral oscillation peaks separated by 0.07 nm, indicating self-sustained pulsations. The corresponding RIN spectrum shows an increased noise intensity at a frequency of 9 GHz. Coherence collapse is observed at R = 12 dB for the IC-DFB laser. The results for the CCDFB laser differ significantly from the IC structures: No self-sustained pulsations are observed and the emission remains stable and single-mode (line at 1542.3 nm) even for strong feedback R. The laser is insensitive to optical feedback up to R = 9 dB, where the coherence collapse sets on. For both device types the laser linewidth is strongly broadened in the collapse regime and a maximum RIN value > 120 dB/Hz is measured. Increased DFB currents shift the onset of coherence collapse toward higher feedback strengths for both types. All together, the CC-lasers exhibit

Proc. of SPIE Vol. 5956 59560I-5
significantly higher robustness against feedback and operate single mode up to R = 9 dB, while IC-lasers start to degrade for R = 24 dB already.

R=-11dB.-8dB

feedback R (dB)

RIN (dB/Hz)

-16 -20 -24 -28 -32

coherence collapse

R=-8dB

-130 -140

R=-18dB.-12dB

-130 -140 -150

R < -9dB

opt. power (dBm/0.01nm)

-25 -45 -65 1554.40

onset of degradation

R<-19dB

1555.90

1542.15 1542.45 1541.75

1554.90

1541.85

wavelength (nm)

frequency (GHz)
Fig. 8: Optical spectra (left side), color coded, for various feedback strengths (y-axis) and RIN spectra (right side) for IC-DFB laser (a) and CC-DFB laser (b). Lasers were driven 60 mA above threshold.
The results obtained under dc operation were complemented by investigating the robustness against optical feedback in a 2.5 Gbit/s direct modulation experiment. The lasers were operated at 50 mA above threshold. Via a bias-T the lasers were additionally modulated by a PRBS 27-1 NRZ data signal. The same feedback setup as for the dc experiment was used, however, instead of measuring the spectra, the signal quality of the 2.5 Gbit/s data signal was determined by BER measurements.
IC - D F B : fille d s y m b o ls C C -D F B : o p e n s ym b o ls
R < -d B R = -d B R = -8.2 d B

log (BER)

3.5 d B
----o p tic a l p o w er (d B m ) -2 8
Fig. 9: Comparison of the BER at 2.5 Gbit/s for IC-DFB and CC-DFB lasers under various feedback strengths R. The laser currents were chosen at 50 mA above threshold (cf. Fig. 8).
Figure 9 summarizes the BER performance for both laser types. Without feedback (R < 40 dB) both devices show the same BER characteristics. The BER performance of the IC-DFB laser starts to degrade at a feedback strength of about 20 dB with a power penalty of 2 dB, which corresponds to optimized IC-DFB lasers [16]. For 13 dB feedback strength a significant penalty is already observed for the IC-DFB laser, while the performance of the CC-DFB laser is not yet affected, and in the collapse regime (BER curves for maximal feedback strength of 8 dB) both lasers show a penalty, however, smaller by 3.5 dB for the CC-DFB type.
Proc. of SPIE Vol. 5956 59560I-6

2.6. ps OEICs Mode-locked lasers (MLL) as optical pulse sources have been designed for a number of applications, such as: 1) ultrahigh bit rate optical telecom systems (> 40 Git/s) utilizing optical time division multiplexing (OTDM) transmission techniques, 2) optical signal processing (e.g. switching, regeneration, analogue/digital conversion, clock distribution), 3) radio over fibre communication systems, and 4) ultrahigh speed measurement equipment (e.g. optical sampling, testing of optoelectronic devices, pump-and-probe measurements). Semiconductor mode-locked lasers, based on hybrid or monolithic integration [e.g. 17, 18], are promising solutions for such applications. In particular, monolithic mode-locked diode lasers are attractive due to their compactness, stability, reliability, and economic fabrication potential. In many cases very demanding specifications have to be fulfilled with respect to pulse width, amplitude noise and timing jitter, and for high speed optical communication, e.g. in 160 Gbit/s OTDM systems, requirements are < 2 ps, < 3 % and < 300 fs, respectively. In particular, amplitude noise caused by Q-switched mode-locking (QML) is a challenge, and this is of particular relevance, if very short optical pulses and low amplitude noise at high repetition rates have to be achieved simultaneously [19]. Amplitude noise in mode-locked lasers with integrated saturable absorber (SA) is to a significant extent due to Qswitched mode-locking, and a substantial suppression of QML is expected from an optimisation of the saturation behaviour of the saturable absorber and/or the gain section. One means to achieve this goal is to increase the saturation energy Esat as much as possible, and as Esat is inversely proportional to the number of QWs it is expected that the lower the number of QWs the better device characteristics with respect to pulse width and amplitude noise can be achieved. We have fabricated monolithic, four-section mode-locked lasers, comprising SA-, gain-, phase-, and DBR-sections with structural and fabrication details as described elsewhere [20], and we have made a detailed comparison of the characteristics of MLLs comprising three and six QWs [21]. The devices have been operated with a large number of different combinations of gain current and SA-voltage, and the corresponding values of amplitude noise and pulse width are compiled in Fig. 10.

Amplitude noise / %

QW QW 5
Fig. 10: Amplitude noise vs. pulse width for 3-QW- and 6QW mode-locked lasers.

Pulse width / ps

It is obvious from that figure that on the one hand the 3 QW devices are superior to the 6 QW ones, and that <2 ps pulse width and <3 % amplitude noise are obtained for a large number of operation parameters.
2.7. High speed Mach-Zehnder modulators for on-off-keying and differential phase-shift keying Modulators based on a Mach-Zehnder interferometer (MZI) structure are the first choice for long-haul transmission [22]. Such MZI-type modulators can not only be used for on-off keying, but they exhibit excellent performance as well, if they are used for phase modulation schemes like differential phase shift keying (DPSK) and return-to-zero (RZ)-DPSK, which have received increased attention recently [23]. Based on InP and designed with capacitively loaded travelling wave electrodes (TWEs) (cf. Fig. 11) such devices are well suited for high-speed transmission at 40 Gbit/s and beyond. Moreover, they have the potential for monolithic integration with a tuneable laser diode. In order to reduce the packaging cost by improved alignment tolerances and in order to enhance the reliability and to decrease the insertion loss, it is essential to integrate a spot size converter (SSC), which enables efficient fibre/chip coupling [24]. A corresponding layout is shown in Fig. 12. In order to perform the vertical guiding in the fibre-port waveguide, three thin quaternary layers separated by InP are added beneath the MQW layers.
Proc. of SPIE Vol. 5956 59560I-7
High data rate operation is assured by the TWEs, which are designed as coplanar lines, where the overall impedance can be matched to 50 by the capacitive load of the distributed electrodes on the MZ arms. The electro-optical (EO) response of this device is shown in Fig. 13. The optical output signal is transferred to the electrical domain using a high speed photodiode (u2t Photonics AG) and is measured using an electrical spectrum analyser. It shows a 2 dB bandwidth at 50 GHz, which is the limit of the actual measuring setup. The corresponding 3 dB bandwidth is estimated to be 58 GHz based on electrical S 21 parameter measurements, which are carried out up to 70 GHz. High frequency device simulations demonstrate additional optimisation potential up to 90 GHz 3-dB bandwidth.

Z = 50W

evi ssap
Z = Z = 50 W without loading

active

RF input

WG 50 c

output

gn idiug

Fig. 11: Schematic layout of an MZI based modulator with capacitively loaded travelling wave electrodes.

Fig. 12: Layout of the spot size converter; field distribution for the fundamental mode in the passive and fibre port waveguide, respectively.
Fig. 13: Electro-optical (EO) response and electrical S21-parameter.
A packaged modulator for 40 Gbit/s is shown in Fig. 14. Actual developments are aiming towards monolithic integrated transmitters for DPSK and DQPSK with transmission rates up to 100 Gbit/s.
3. PHOTODETECTORS AND RECEIVERS
3.1. pin-TWA photoreceiver InP-based opto-electronic integrated receivers are particularly suited for 1.55 m transmission wavelength, high-bit rate systems of 80 Gbit/s and beyond, and they are of high interest for RF and microwave instrumentation as well. Current development goals include increased component efficiency, further miniaturization, and cost effective packaging by raising the degree of integration, and corresponding recent achievements will be discussed in the following section.
e l e c t r i c a l S [ d B ]

] z H G[ y cn e uq erf

1 2S l acirt cel e

es n op ser -O E

0 -1 -2 -3 -4 -5 -6
Fig. 14: Packaged MZ-Modulator for 40 Gbit/s.
Proc. of SPIE Vol. 5956 59560I-8

ediugeva w

t rop-erbif

ediug evaw

srey al
E O r e s p o n s e [ d B ]
The concept of combining a waveguide-integrated photodiode with a spot-size converter and a HEMT-based amplifier and allowing the independent optimization of each of the components has been widely discussed in earlier publications [25, 26]. For 80 Gbit/s system applications a new circuit design has been developed, using HEMTs with modified layer structures and shorter gate lengths. The biasing configuration is accomplished with a negative bias supply at the common source electrode with a metal-insulator-metal (MIM) capacitor [27]. The circuit diagram of the photoreceiver with a negative bias and ground (GND)-isolated output port is depicted in Fig. 15. The amplifiers dc current is fed into the terminal Vdd. The HEMTs (gate length 0.18 m) exhibit maximum cut-off frequencies fT/fmax of typically 140/300 GHz.
Fig. 15: Circuit diagram of the photoreceiver with negative bias and GND-isolated output port.
The simulated transimpedance ZT and S 22 of the travelling wave amplifier (TWA) are illustrated in Fig. 16. The 3-dB bandwidth and transimpedance (ZT) are 65 GHz and 40 dB, respectively. The reflection coefficient S22 is less than 10 dB over almost the whole frequency range up to 80 GHz. The photoreceiver is composed of a waveguide-integrated photodiode with a spot-size converter and a special HEMT-based travelling wave amplifier, and the input facet of the photoreceiver is antireflection-coated using TiO2/SiO2 layers. Additional details of the fabrication process are reported elsewhere [28]. A partial view of the final

!._.re.

hotodiode
Fig. 16: Simulated ZT and S22 of the travelling wave amplifier.
travelling wave amplifier
Fig. 17: Partial view of integrated photoreceiver OEIC.
Proc. of SPIE Vol. 5956 59560I-9

at TWA floating GND

)zHG( ycneuqerf
S ( d B ) Z, T ( d B O h m )

,.j. JiT1.1L

chip is given in Fig. 17, which shows the photodiode at the left being connected to the input of the travelling wave amplifier via an air bridge. The on-wafer high-frequency measurements of the integrated distributed amplifier within the monolithic photoreceiver OEIC were performed with a network analyzer in the frequency range of 0.45 to 110 GHz. The amplifier characteristics defined by the transimpedance (ZT) are derived from the measured S-parameters. The transimpedance amounts to 39 dB (71) and has a bandwidth of 72 GHz. The packaged photoreceiver is shown in Fig. 18. The optical coupling inside the module is done by fixing the butt fibre directly at the chips waveguide facet by UV curable resin. The output of the chip is connected to a 1 mm coaxial connector via a short coplanar waveguide (CPW) on a quartz substrate to fit the geometry of the connector and has a bandwidth over 110 GHz. On the upper side of the module is the dc input.

OOOO)F)O)

normal, power (dB)
Fig. 18: A photograph of a pig-tailed OEIC module with 1 mm RF output.
Fig. 19: Opto-electronic frequency response of the photoreceiver module.
The frequency characteristics of the module are shown in Fig. 19. The 3 dB bandwidth exceeds 70 GHz, which is comparable to the OEIC characteristics and proves a high quality RF packaging. The photodiode dc responsivity amounts to 0.66 A/W with a polarization dependent loss of 0.87 dB. The overall conversion gain of the photoreceiver module is 45.4 V/W, which is high enough for 80 Gbit/s electrical time division multiplexing (ETDM) systems as well as for high frequency measurement equipment. For the time domain characterisation of the modules 80 Gbit/s return-to-zero (RZ) modulated data streams (PRBS) with 2 7-1 pattern length have been generated from 1.7 ps wide pulses of a model-locked semiconductor laser with 10 GHz repetition frequency. Eye diagrams for +7 dBm input power and measured with a 70 GHz sampling scope (Agilent 86100B with Agilent 86118A) are shown in Fig. 20. The detected electrical pulse had a FWHM of 6.8 ps and an under-shoot of less than 10%, including the bandwidth limitation of the sampling scope. The eye diagram is clearly opened with negligible saturation effects [29]. The operating conditions for the measurement setup are: Vs = 2 V, Vdt = 1.89 V, Idt = 37 mA, and Vpn = 0 V.
Fig. 20: 80 Gbit/s RZ eye pattern at +7 dBm optical input power detected by the receiver module.
Proc. of SPIE Vol. 5956 59560I-10
3.2. 100 GHz photodetector A continuous optimisation process has recently enabled the fabrication of photodetector chips with 107 GHz 3-dB bandwidth, for further details about the device design and performance, see Ref. [30]. Chips have been assembled into a standard housing equipped with fibre pigtail and 1 mm coaxial output connector [31], and Fig. 21 shows the frequency response of the fabricated PD module, which reveals a 3-dB bandwidth of 100 GHz, in close agreement with the chip-based measurement.

0 rel. power [dB]

Fig. 21: Relative frequency response of the PD module measured at +2.3 dBm optical input power.

: 1.55 m Vbias: 2 V

80 frequency [GHz]
The dc responsivity of the PD module amounts to 0.63 A/W with a polarisation-dependent loss (PDL) of only 0.64 dB at 1.55 m wavelength. Eye diagrams at 80 Gbit/s data rate have been measured using the experimental setup described in [32]. Corresponding results for +8 dBm and +11 dBm input power are shown in Fig. 22. For the latter case the average photocurrent is 8 mA, and the eye opening exceeds 0.5 V peak voltage at a moderate reverse bias of 2.5 V. This value is twice as much as required by state-of-the-art demultiplexing electronics.
Fig. 22: 80 Gbit/s RZ eye pattern at +8 dBm (left) and +11 dBm (right) optical input power detected by the photodetector module at 2.5 V reverse bias, PRBS = 27-1.
3.3 Balanced photoreceiver Balanced photoreceivers and detectors have received high interest recently as key components in differential (quadrature) phase-shift keying (D(Q)PSK)-systems, which offer high sensitivity and robustness in long haul fibre links [33]. Monolithically integrated balanced photodetectors provide a particular high common-mode rejection ratio, broad bandwidth and reduced packaging cost [34]. If these devices contain a 50 matching resistance at the joined output of the two photodiodes an especially low electrical output reflection is obtained, although at the expense of loosing 50 % of the signal level, equivalent to 6 dB loss of the electrical power. A monolithic integration of an InP-based amplified balanced photoreceiver OEIC, comprising two waveguide-integrated photodiodes and a broadband distributed amplifier is shown in Fig. 23. The integration of the distributed amplifier provides about 12 dB of additional power gain and ensures a low output reflection, guaranteeing good signal integrity in conjunction with subsequent electronics.
Proc. of SPIE Vol. 5956 59560I-11

c lers

spot size

waveguides

Fig. 23: Micrograph of the balanced photoreceiver chip.

4. SUMMARY

Examples of lasers, which enable low cost telecommunication systems, have been presented, including lasers for uncooled operation and direct modulation at 10 Gbit/s, complex-coupled lasers with low sensitivity to back reflections, and lasers for efficient coupling to optical fibres. Other InP-based devices of high current interest are monolithically integrated mode-locked lasers as ps pulse sources, high bit rate Mach-Zehnder interferometer modulators, ultra high speed photodetectors (> 100 GHz bandwidth) and integrated photoreceivers, which represent building blocks for next generation high speed and robust networks.

REFERENCES

1. 2. J. Buus, M.-C. Amann, and D. Blumenthal: Tunable Laser Diodes an Related Optical Sources, 2 nd ed. (Wiley-IEEE Press, New York, 2005) Y.A. Akulova, G.A. Fish, P. Koh, P. Kozodoy, M. Larson, C. Schow, E. Hall, H. Marchand, P. Abraham, and L.A. Coldren: "10 Gb/s Mach-Zehnder modulator integrated with widely-tunable sampled grating DBR laser", Techn. Digest Optical Fiber Commun. Conf. (OFC 2004), Los Angeles, CA, USA, paper TuE4 D.J. Robbins, G. Buscio, L. Ponnampalam, J.P. Duck, P.J. Williams, R.A. Griffin, A.J. Ward, D.C.J. Reid, N.D. Whitbread, and E. Barton: "A high power, broadband tuneable laser module based on a DS-DBR laser with integrated SOA," Techn. Digest Optical Fiber Commun. Conf. (OFC 2004), Los Angeles, CA, USA, paper TuE3 J.-O. Wesstrm, G. Sarlet, S. Hammerfeldt, L. Lundqvist, P. Szabo, and P.-J. Rigole: "State-of-the-art performance of widely tunable modulated grating Y-branch lasers", Techn. Digest Optical Fiber Commun. Conf. (OFC 2004), Los Angeles, CA, USA, paper TuE2 H. Ishii, H. Oohashi, K. Kazuo, K. Tsuzuki, and Y. Tohmori: "High-power (40 mW) L-band tunable DFB laser array module using current tuning," Techn. Digest Optical Fiber Commun. Conf. (OFC 2005), Anaheim, CA, USA, paper OTuE1 R. Paoletti, M. Agresti, D. Bertone, L. Bianco, C. Bruschi, A. Buccieri, R. Campi, C. Dorigoni, P. Gotta, M. Liotti, G. Magnetti, P. Montangero, G. Morello, C. Rigo, E. Riva, D. Soderstrom, S. Stano, P. Valenti, M. Vallone, and M. Meliga: "Highly reliable and high yield 1300 nm InGaAlAs directly modulated ridge Fabry-Perot lasers, operating at 10 Gb/s, up to 110 C, with constant current swing," Techn. Digest Optical Fiber Commun. Conf. (OFC 2005), Anaheim, CA, USA, postdeadline paper PDP15 T. Nakamura and N. Suzuki: "Spot-size converted laser diodes based on mode interference," IEEE Photon. Technol. Lett. 12, 143-145 (2000) A. Uda, K. Tsuruoka, N. Suzuki, K. Fukushima, T. Nakamura, and T. Torikai: "Spot-size Expanded High Efficiency 1.3 m MQW Laser Diodes with Laterally Tapered Active Stripe," Proc. 9 th IPRM, Hyannis, USA, pp. 657-660 (1997) M. Mhrle, H. Roehle, A. Sigmund, A. Suna, and F. Reier: "High-performance all-active tapered 1550 nm InGaAsP-BH-FP lasers," Proc. 14th IPRM, Stockholm, Sweden, pp. 27-30 (2002) M. Kito, Y. Inaba, H. Nakayama, T. Chino, M. Ishino, Y. Matsui, and K. Itoh: High Slope Efficiency and Low Noise Characteristics in Tapered-Active-Stripe DFB Lasers with Narrow Beam divergence," IEEE J. Quantum Electron. 35, 17651770 (1999) M. Mhrle, A. Sigmund, R. Steingrber, W. Frst, and A. Suna: "All-active tapered 1.55 m InGaAsP-BH-DFB-laser with continously chirped grating," IEEE Photon. Technol. Lett. 15, 365-367 (2003) M. Mhrle, to be published at ECOC 2005 K. Petermann: "External Optical Feedback Phenomena in Semiconductor Lasers IEEE," J. Select. Topics Quantum Electron. 1, 480-489 (1995) J. Kreissl, W. Brinker, E. Lenz, T. Gaertner, W. Rehbein, S. Bauer, and B. Sartorius: "Isolator-free directly modulated complex-coupled DFB lasers for low cost applications," Techn. Digest Optical Fiber Commun. Conf. (OFC 2005), Anaheim, CA, USA, paper OThU3 J. Kreissl, U. Troppenz, W. Rehbein, B. Httl, E. Lenz, and H. Venghaus: "High single-mode-yield multiple-wavelength DFBlaser arrays in the 1.55 m range," Proc. 27th Europ. Conf. Opt. Commun. (ECOC 2001), Amsterdam, paper TU B 1.1, pp. 118

7. 8. 9. 10. 11. 12. 13. 14. 15.
Proc. of SPIE Vol. 5956 59560I-13
16. F. Grillot, B. Thedrez, J. Py, O.Gauthier-Lafaye, V. Voiriot, and J.L. Lafragette: "2.5-Gb/s Transmission Characteristics of 1.3-m DFB Lasers with External Optical Feedback," IEEE Photon. Technol. Lett. 14, 101-103 (2002) 17. R. Ludwig, S. Diez, A. Erhardt, L. Kller, W. Pieper, and H.G. Weber: "A tunable femtosecond modelocked semiconductor laser for applications in OTDM-systems," IEICE Trans. Electron. E81-C, 140-145 (1998) 18. E.A. Avrutin, J.H. Marsh, and E.L. Portnoi: "Monolithic and multi-GigaHertz mode-locked semiconductor laser: Constructions, experiments, models and applications," IEE Proc.-Optoelectron. 147, 251-278 (2000) 19. U. Keller: "Recent developments in compact ultrafast lasers," Nature 424, 831-838 (2003) 20. B. Httl, R. Kaiser, Ch. Kindel, S. Fidorra, W. Rehbein, H. Stolpe, G. Sahin, U. Bandelow, M. Radziunas, A. Vladimirov, and H. Heidrich: "Reduction of amplitude noise in monolithic MQW mode-locked DBR lases on GaInAsP/InP," IEEE Photon. Technol. Lett. (in press) 21. B. Huettl, R. Kaiser, W. Rehbein, H. Stolpe, Ch. Kindel, S. Fidorra, A. Steffan, A. Umbach, and H. Heidrich: "Low noise monolithic 40 GHz mode-locked DBR lasers based on GaInAsP/InP," Proc. 17th Int. Conf. on InP and Related Materials (IPRM 2005), Glasgow, Scotland, 2005 22. Ed L. Wooten, K.M. Kissa, A. Yi-Yan, E.J. Murphy, D.A. Lafaw, P.F. Hallemeier, D. Maack, D.V. Attanasio, D.J. Fritz, G.J. McBrien, and D.E. Bossi: "A review of Lithium Niobate modulators for fiber-optic communications systems," IEEE J. Select. Topics Quantum Electron. 6, 69-82, (2000) 23. A. H. Gnauck, S. Chandrasekhar, J. Leuthold, and L. Stulz: "Demonstration of 42.7 Gb/s DPSK receiver with 45 photons/bit sensitivity," IEEE Photon. Technol. Lett. 15, 99-101 (2003) 24. D. Hoffmann, S. Staroske, and K.-O. Velthaus: "45 GHz bandwidth travelling wave electrode Mach-Zehnder-modulator with integrated spot size converter," Proc. 16 th Int. Conf. on InP and Related Materials (IPRM 2004), 585-588 (2004) 25. H.-G. Bach, A. Umbach, S. van Waasen, R.M. Bertenburg, and G. Unterbrsch, "Ultrafast monolithically integrated InP-based photoreceiver: "OEIC-design, fabrication, and system application," Special Issue of IEEE J. Select. Topics Quantum Electron. and Integrated Optics (JSTQE) 2, 418-423 (1996) 26. W. Schlaak, G. G. Mekonnen, H.-G. Bach, C. Bornholdt, C. Schramm, A. Umbach, R. Steingrber, A. Seeger, G. Unterbrsch, W. Passenberg, and P. Wolfram: "40 Gbit/s eyepattern of a photoreceiver OEIC with monolithically integrated spot size converter," Techn. Digest Optical Fiber Commun. Conf. (OFC 2001), Anaheim, CA, USA, paper WQ4, pp. 1-3 27. G.G. Mekonnen, H.-G. Bach, W. Schlaak, R. Steingrber, A. Seeger, W. Passenberg, W. Ebert, G. Jacumeit, Th. Eckhardt, R. Ziegler, and A. Beling: "40 Gbit/s Photoreceiver with DC-Coupled Output and Operation without Bias-T," Proc. 14th Intern. Conf. on InP and Related Materials (IPRM 2002), May 12-16, 2002, Stockholm, Sweden, paper A8-1 28. G.G. Mekonnen, W. Schlaak, H.-G. Bach, R. Steingrber, A. Seeger, Th. Engel, W. Passenberg, A. Umbach, C. Schramm, G. Unterbrsch, and S. van Waasen: "37-GHz Bandwidth InP-Based Photoreceiver OEIC Suitable for Data Rates up to 50 Gb/s," IEEE Photon. Technol. Lett. 11, 257-259 (1999) 29. G.G. Mekonnen, H.-G. Bach, A. Beling, R. Kunkel, D. Schmidt, and W. Schlaak: "80-Gb/s InP-based waveguide-integrated photoreceiver," IEEE J. Select. Topics Quantum Electron. 11, 356-360 (2005) 30. H.-G. Bach, A. Beling, G. G. Mekonnen, R. Kunkel, D. Schmidt, W. Ebert, A. Seeger, M. Stollberg, and W. Schlaak: " InPBased Waveguide-Integrated Photodetector with 100 GHz Bandwidth," IEEE J. Select. Topics Quantum Electron. 10, 668-672 (2004) 31. A. Beling, H.-G. Bach, G. G. Mekonnen, T. Eckhardt, R. Kunkel, G. Jacumeit, M. Kroh, and J. Berger: "Fully Packaged InPbased Photodetector for 80/85 Gbit/s RZ Systems," Proc. 29 th Europ. Conf. Opt. (ECOC 2003), Rimini, Italy, Vol. 6, 82-83 (2003) 32. A. Beling, H.-G. Bach, G.G. Mekonnen, T. Eckhardt, R. Kunkel, D. Schmidt, and C. Schubert: "Highly Efficient PIN Photodetector Module for 80 Gbit/s and Beyond," Techn. Digest Optical Fiber Commun. Conf. (OFC 2005), Anaheim, CA, USA, paper OFM1 33. A.H. Gnauck, S. Chandrasekhar, J. Leuthold, and L. Stulz: "Demonstration of 42.7 Gb/s DPSK receiver with 45 photons/bit sensitivity," IEEE Photon. Technol. Lett. 15 (1), 99-101 (2003) 34. A. Beling, H.-G. Bach, D. Schmidt, G.G. Mekonnen, M. Rohde, L. Molle, H. Ehlers, and A. Umbach: "High-Speed Balanced Photodetector Module with 20dB Broadband Common-Mode Rejection Ratio," Techn. Digest Optical Fiber Commun. Conf. (OFC 2003), Atlanta, GA, USA, vol. 1, 339-340 (2003) 35. C. Schramm, H.-G. Bach, A. Umbach, A. Beling , G. Jacumeit, S. Ferber, R. Ludwig, R. Ziegler, G.G. Mekonnen, R. Kunkel, D. Schmidt, W. Schlaak, and G. Unterbrsch: "High-Bandwidth Balanced Photoreceiver Suitable for 40-Gb/s RZ-DPSK Modulation Formats," Special Issue of IEEE J. Select. Topics Quantum Electron. and Integrated Optics (JSTQE) 11, 127-134 (2005) 36. H.-G. Bach, W. Schlaak, G. Unterbrsch, G.G. Mekonnen, G. Jacumeit, R. Ziegler, R. Steingrber, A. Seeger, Th. Engel, A. Umbach, C. Schramm, and W. Passenberg: "40 Gbit/s Photoreceiver Modules Comprising InP-OEICs for RZ and NRZ Coded TDM System Applications," Techn. Digest, OFC99 / IOOC, Feb 21-26, 1999, San Diego, USA, paper TuI5, pp. 108-110 37. M.S. Islam, T. Chau, S. Mathai, T. Itoh, M.C. Wu, D.L. Sivco, and A.Y. Cho: "Distributed Balanced Photodetectors for Broad-Band Noise Suppression," IEEE Trans. Microwave Theory Tech. 47, 1282-1287 (1999) 38. J.H. Sinsky, A. Adamiecki, A. Gnauck, C.A. Burrus, S. Chandrasekhar, J. Leuthold, and O. Wohlgemuth: "A 40-Gb/s Integrated Balanced Optical Front End and RZ-DPSK Performance," IEEE Photon. Technol. Lett. 15, 1135-1137 (2003)