1. 1 Broad-band and Scalable Circuit-level Models of MSM PD for Co-design with Preamplifier in Front-end Rx Applications Ph.D. Defense Spring, 2004 Cheol-ung Cha Advisor: Prof. Martin A. Brooke School of Electrical and Computer Engineering Georgia Institute of Technology Atlanta, GA, 30332

2. 2 Outline  Optical Interconnects and Communications  MSM Photodetector  Preamplifier  Modeling Methodology  Motivation  Previous Modeling Work  Partial Element Equivalent Circuit (PEEC) Model  Proposed Modeling Method  Partial Elements (PEs) and Test structures  Measurement-based PEEC (M-PEEC) Model  Modeling Procedure  Case study: Straight Line Modeling  Calibration  On-wafer Calibration  MSM Photodetector Modeling  Partial Elements (PEs) and Test structures  M-PEEC Model Extraction  Optimization Results  Conclusions

3. 3 Optical Interconnects & Communications  General OIC system Transmission channel Tx Power Control Electrical Signal Laser Driver Re-timer PLL Serializer MUX AGC Recovered Signal Decision Circuitry Clock Recovery Deserializer Rx TIA Limiting Amp. DEMUX

4. 4  Metal-Semiconductor-Metal Photo-Diode (MSM PD)  Role : Optical signal Electrical signal  Condition : hv >Eg (optical) and reverse voltage bias (electrical) MSM PD: What is MSM PD? Interdigitated fingers Contact pads Frame A B +-- E Lightwave EE

5. 5  Advantages of MSM PD  Low capacitance  Broad bandwidth  Ease of monolithic integration with FETs  Ease of alignment  Low dark current (~ nA scale)  Drawback of MSM PD  Low responsivity (about 0.2~0.4) (Low output current level requires sensitive preamplifier design)  FWHM : 12.46ps MSM PD: Advantages

6. 6 MSM PD: Capacitance  Capacitance is  Major parasitic component of the MSM PD  Main limitation factor for high-frequency (multi-GHz) applications  Three times smaller than that of PIN PD (Large detection area enables higher alignment tolerance for packaging)  Conventional formulas are based on the Microwave theory  Obtained without illumination of light  Obtained without considering frame Ex) Conformal mapping theory only considered interdigitated fingers without considering the effects of frame and light illumination.

7. 7 MSM PD: Capacitance  Simulation results with preamplifier with respect to different capacitance values : 50, 80, 100 fF 3dB 2.2 3 4.2 50 fF 80 fF 100 fF

8. 8 MSM PD: Capacitance Pad MSM PD w/ & w/o illumination of light Frame

9. 9 MSM PD: Capacitance  Comparison of measured S22

10. 10 MSM PD: Capacitance  Interdigitated fingers: Conformal mapping theory where,, Depends on size, finger width, and spacing  Frames: Complete elliptic integral of the second kind Where,, and

11. 11 MSM PD: Capacitance  Light illumination: External quantum efficiency where and  Total capacitance

12. 12 MSM PD: Capacitance 18 fF18.2 fF Superposition (C Total ) By subtraction 3 fF By proposed formula 2.7 fF Capacitance from illumination of light (C light ) 6 fF By proposed formula 5.5 fF Capacitance of Frames (C frame ) NA By conformal mapping 10 fF Capacitance of interdigitated electrodes (C fingers ) MeasurementTheory What makes this huge difference?  20/1/2 MSM photodetector

13. 13 MSM PD: Transit time & BW  Transit time  The time for a carrier to take to travel through the active region and collected by contacts.  Low mobility of hole causes a long tail in the impulse response and small bandwidth in the frequency response.  The transit time is where is the saturated carrier velocity and d is the distance of travel. Depends on finger spacing  Bandwidth (BW)  Two main factors that limit the speed is “capacitance” and “transit time”  Trade off between capacitance and transit time (size, finger spacing, and width).  The BW is RC time const. Transit time const.

14. 14  Bandwidth of Square MSM PDs 3dB freq. of transit time const. 3dB freq. of RC time const. 3dB freq. of total time const. MSM PD: Bandwidth

15. 15  Total bandwidth of MSM PDs (Trade off between RC and transit time const.) 20x20 MSM PDs 40x40 MSM PDs 60x60 MSM PDs 80x80 MSM PDs MSM PD: Bandwidth

16. 16 MSM PD: Lumped Equivalent-circuit Model  Equivalent-circuit model of pad and MSM PD 10 Gbps 20 Gbps 30 Gbps

17. 17 Preamplifier: Performance Metrics  Key performance metrics of optical receiver  Bandwidth, Sensitivity, Noise, and Gain  Mainly determined by front-end (preamplifier and photodetector)  TransImpedance Amplifier (TIA)  Convert low-level photocurrent to usable voltage signal  Feedback in preamplifier  Extending BW  Reducing noise (Good sensitivity)  Controlling input and output impedance  The close-loop gain is where the open-loop gain. β xsxs xfxf xoxo + - Ao(ω)

18. 18  MSM PD with commercial TIA ( Maxim 2.5 Gbps TIA) Preamplifier: Eye Diagrams BERT Oscilloscope 50 Ohm Matched 1.2 Gbps 2 Gbps Modulator Laser 2.5 Gbps TIA MSM PD 6Gbps3Gbps The output current of MSM PD (60/1/2) is too weak to be detected by oscilloscope

19. 19 Outline  Optical Interconnects and Communications  MSM PD  Preamplifier  Modeling Methodology  Motivation  Previous Modeling Work  Partial Element Equivalent Circuit (PEEC) Model  Proposed Modeling Method  Partial Elements (PEs) and Test structures  Measurement-based PEEC (M-PEEC) Model  Modeling Procedure  Case study: Straight Line Modeling  Calibration  On-wafer Calibration  MSM Photodetector Modeling  Partial Elements (PEs) and Test structures  M-PEEC Model Extraction  Optimization Results  Conclusions

20. 20 Motivation: Higher Performance  Demand for higher bandwidth and speed requires well-designed front-end (preamplifier with photodetector) of optical Rx.  Front-end is a dominant component in a Rx because the sensitivity of the Rx is mainly determined by the noise factor of the front-end.  Reduction in bandwidth comes from the parasitic capacitance of a photodetector and pad. The capacitance of bond-pad is typically 10–50 fF (significant for GHz circuitry). - Flip-chip bonding techniques can be used to reduce parasitics at the interface between InGaAs and CMOS. The capacitance of commercial PIN and avalanche photodiode is 200–900 fF. - Using MSM PDs, this value can be reduced up to 50-300 fF. (The reduced capacitance would allow enough budgets for circuit design), Solution Co-design of photodetector with preamplifier is a solution : when a circuit designer design circuitry, he/she can choose proper device specifications such as device size, finger spacing and width, and thickness of active layer to satisfy the requirements.

21. 21 Motivation: Modeling Method  Modeling methodology for co-design should be  Easy to use (Needs to be integrated into existing circuit design environment such as HSPICE and ADS. - This approach circumvents the inconvenient, iterative interface between a photonic device simulator and a circuit design tool.  Fast - The finite-element methods need long simulation time and huge memory resource  Accurate - Existing analytical equation-based methods are not accurate.  Scalable - Modeling method can predict the model of different dimensional device.

22. 22 Modeling Methodology Tree Time domain Frequency domain Analytical (Equation-based) Finite Element (Spatial discretization) Finite Difference Time Domain Partial Element Equivalent Circuit (Discrete Approx. of EFIE) Electric Field Integral Equation Empirical (Measurement-based) Measurement-based Partial Element Equivalent Circuit (M-PEEC) Differential equation (Grids on whole area) Integral equation (Grids only on conductors) Finite Methods (Discretization) Finite Element Equivalent Circuit Method of Moments (MoM) Numerical (EM full wave-based) Proposed in this research Improved in this research for the capacitance modeling of the MSM PD

23. 23 Previous Modeling Work  Earlier work in high frequency component modeling mainly originated from the microwave engineering community.  Three fundamental methodologies  Analytical equation-based modeling method Direct derivation from first physical principles - very few, available only for very simple structures Generally difficult and time consuming to develop Not very flexible Not accurate  Numerical EM-full wave based modeling method Accurate Highly flexible Very slow and requiring huge memory resource, so it’s not practical for complex geometry system analysis

24. 24 Two dominant methods exist (continued) - The Finite Element Method (FEM) FEM yields high accuracy for 3 dimensional structures. Grids structure into many small pieces, and solves Maxwell’s Equations - The Method of Moment (MoM) MoM is a 2 1/2-D method with less accuracy in 3 dimensions. Assumes a conductor height of zero. Grids structure into many small pieces, and solves Green’s Function  Measurement-based modeling method Measured data from time or frequency domain can be fit to a circuit model using optimization techniques Non-ideal processing effects can be considered The method allows for statistical modeling Very accurate for measured structures Not very flexible Previous Modeling Work Improved measurement-based, scalable, and flexible modeling method

25. 25 Partial Element Equivalent Circuit (PEEC) Model  Three dimensional partial element equivalent circuit (PEEC) model was originated from high-speed interconnect modeling in 1970s[Ruehli].  The PEEC method is based on Maxwell’s integral equation that is interpreted in terms of RLC elements and their couplings.  Maxwell’s Electric Field Integral Equation (EFIE)  The advantages of the PEEC method are  The output of the PEEC analysis is spice-like equivalent-circuit model (it can be easily integrated with other circuit models such as transistor models into a conventional circuit simulation tools such as SPICE).  The PEEC models work equally well in the time and frequency domains.  The PEEC analysis can reduce simulation time by using Maxwell’s integral equation.  The PEEC models include cross coupling terms.

26. 26 Partial Element Equivalent Circuit (PEEC) Model Pad Partial Element (PE)Square Partial Element (PE) CS 15 CS 13 CS 35 LS 22 RS 22 RS 44 LS 44 CS 11 LS 24 CS 55 CS 33 CP 15 CP 13 CP 35 LP 22 RP 22 RP 44 LP 44 CP 11 LP 24 CP 55 CP 33

27. 27 Partial Element Equivalent Circuit (PEEC) Model  Primitive PEEC cell Capacitive cell_1Capacitive cell_5Capacitive cell_3 Inductive cell_2Inductive cell_4 w d C 15 C 13 C 35 C 15 L 22 R 22 R 44 L 44 L 24  In the general case, the i th circuit equations of n inductive and m capacitive cells are where and are the index of the capacitive cells connected to inductive cell i.

28. 28 Outline  Optical Interconnects and Communications  MSM PD  Preamplifier  Modeling Methodology  Motivation  Previous Modeling Work  Partial Element Equivalent Circuit (PEEC) Model  Proposed Modeling Method  Partial Elements (PEs) and Test structures  Measurement-based PEEC (M-PEEC) Model  Modeling Procedure  Case study: Straight Line Modeling  Calibration  On-wafer Calibration  MSM Photodetector Modeling  Partial Elements (PEs) and Test structures  M-PEEC Model Extraction  Optimization Results  Conclusions

29. 29  If we can accurately model individual parts of a structure, then we can predictively model any structure comprised of those parts accurately.  Those individual parts are called “Partial Elements” (PEs).  “Test structures” are designed, fabricated, and measured to extract the equivalent circuit models, which are called “ Measurement-based partial element equivalent circuits (M-PEECs).”  Partial elements must have enough sensitivity within a test structure in order to be deembedded.  Initial guesses are derived from measured S-parameters.  Optimized M-PEEC models, which are resulted from one test structure, are used in extracting other M-PEEC models for subsequent test structures.  Models of different geometry structures can be created by combining M-PEEC models of partial elements. Partial Elements (PEs) & Test Structures

30. 30 Measurement-based PEEC (M-PEEC) Model  The M-PEEC models have these advantages:  The M-PEEC models are accurate because they are derived from test structures and measurements that automatically include unexpected processing effects such as processing fluctuations, uneven depositions, and non-ideal material properties.  The M-PEEC models can be generated easily and simulated very quickly in a standard and conventional circuit simulator.  The M-PEEC models can be applicable to both electrical and optical devices (passive and active devices) and interconnects modeling which are electrically long and short structures. (In case of optical devices modeling, iterative and inconvenient interface between optical device and electrical circuit simulators can be overcome).  The M-PEEC models are independent of technology or the process in which the structures are fabricated because changed and modified factors are automatically taken into account in the measurements.  The M-PEEC models are scalable and predictive since equivalent-circuit models of different dimensional devices can be constructed from obtained several M-PEEC models.  The M-PEEC models can take into account statistical information in the models.

31. 31  Design and Modeling Flow Modeling Procedure Design & Fab. Test Structures Extract M-PEEC models using optimization Generate Design Rule Library Design Desired Device Co-simulation with Circuitry in SPICE-type Simulator Accurate simulation results Design Rule Checking Pass Fail Define Partial Elements (PEs) Calibration & Measurement What structure to be considered?

32. 32  Straight line is meshed into 20 square PEs and pads by commercial EM simulator (MoM in ADS) Case Study: Straight Line Modeling Coplanar waveguide 20 square PEs

33. 33 Case Study: Straight Line Modeling  Straight line is meshed into 20 square PEs and 2 pads by the proposed modeling method. Pad Partial Element (PE)Square Partial Element (PE) Square M-PEEC Pad M-PEEC

34. 34 Case Study: Straight Line Modeling  Two PEs and their parameter values of M-PEECs Pad Partial Element (PE) Square Partial Element (PE)

35. 35 Case Study: Straight Line Modeling  S11 comparison: measured data, MoM model, and M-PEEC model. Measured data M-PEEC model Mom model

36. 36 Case Study: Straight Line Modeling Measured data M-PEEC model Mom model  S21 comparison: measured data, MoM model, and M-PEEC model.

37. 37 Outline  Optical Interconnects and Communications  MSM PD  Preamplifier  Modeling Methodology  Motivation  Previous Modeling Work  Partial Element Equivalent Circuit (PEEC) Model  Proposed Modeling Method  Partial Elements (PEs) and Test structures  Measurement-based PEEC (M-PEEC) Model  Modeling Procedure  Case study: Straight Line Modeling  Calibration  On-wafer Calibration  MSM Photodetector Modeling  Partial Elements (PEs) and Test structures  M-PEEC Model Extraction  Optimization Results  Conclusions

38. 38 On-wafer Calibration  Calibration : Defining the ends of a measurement system and the begins of a DUT Reference plane

39. 39  SOL on-wafer calibration  SOL (Short-Open-Load)  On-wafer : Calibration structures are on the same substrate with DUT Original Load Trimmed Load Open On-wafer Calibration Short NiCr Resistors

40. 40  Un-trimmed load  Designed for 25 Ohm.  NiCr is used.  Laser-trimmed load  Optimized for 50 Ohm.  NiCr is used. On-wafer Calibration 29.286 Ohm 28.809 Ohm 50.025 Ohm 49.873 Ohm

41. 41 Outline  Optical Interconnects and Communications  MSM PD  Preamplifier  Modeling Methodology  Motivation  Previous Modeling Work  Partial Element Equivalent Circuit (PEEC) Model  Proposed Modeling Method  Partial Elements (PEs) and Test structures  Measurement-based PEEC (M-PEEC) Model  Modeling Procedure  Case study: Straight Line Modeling  Calibration  On-wafer Calibration  MSM Photodetector Modeling  Partial Elements (PEs) and Test structures  M-PEEC Model Extraction  Optimization Results  Conclusions

42. 42  Partial Elements (PEs) and Test structures for MSM PD modeling Partial Elements (PEs) and Test structures Pad PE Line PE Interdigitated PE Test structures

43. 43  “MSM PD” is comprised of “interdigitated partial elements” and couplings Partial Elements (PEs) and Test structures Coupling Capacitance Interdigitated partial element (PE) Coupling Inductance

44. 44 Step I: Pad M-PEEC Model Extraction Pad Extracting Circuit model This obtained “Pad M-PEEC” is used for “Line M-PEEC” extraction. “Line M-PEEC” modeling

45. 45 Step II: Line M-PEEC Model Extraction Line This obtained “Line M-PEEC” is used for “Interdigitated M-PEEC” extraction. Pad Line

46. 46 Step III: Interdigitated M-PEEC Model Extraction Pad This obtained “Interdigitated M-PEEC” is used for “MSM PDs” modeling. Line M-PEEC Interdigitated M-PEEC Line M-PEEC

47. 47 M-PEEC Model Extraction : Parameters  Three PEs and their parameter values of M-PEECs Line Partial Element (PE) Pad Partial Element (PE) Interdigitated Partial Element (PE)

48. 48 Optimization Results: Scalable Model Pad PE Line PE Coupling Inductance Coupling Capacitance Interdigitated PE

49. 49 Optimization Results : Test Structures

50. 50 Optimization Results : Scalable Model  40/1/1 um MSM Photodetector

51. 51 Optimization Results : Scalable Model  40/1/1 um MSM Photodetector

52. 52 Optimization Results : Scalable Model  60/1/1 um MSM Photodetector

53. 53 Optimization Results : Scalable Model  60/1/1 um MSM Photodetector

54. 54 Conclusions  An improved measurement-based modeling method has been proposed and developed for co-design  The main features of developed M-PEEC method are  Accurate  Fast  Scalable and predictive  Process independent  Implementable within existing EDA frameworks such as SPICE  Applicable to 2 and 3-D electrical and optical structures

55. 55 Acknowledgement  Gratitude to:  Dr. Brooke and Dr. Jokerst  Committee members: Dr. Hasler, Dr. Rhodes, Dr. Chang, and Dr. Kohl  Group members

56. 56 Questions and Answers