Methods and Approaches for RF Circuit Simulation And Electromagnetic Modelling

Transcription

1 Methods and Approaches for RF Circuit Simulation And Electromagnetic Modelling T.A.M. Kevenaar 1, E.J.W. ter Maten 1, H.H.J. Janssen 1, S. Onneweer 2 1 Philips Research, Eindhoven, The Netherlands 2 Philips Semiconductors, Sunnyvale, CA, USA 1

2 Overview Introduction RF circuit simulation EM simulation Open problems 2

3 Introduction High Frequency (RF) circuit are gaining importance due to the information society increasing information flows flexible access any time, anywhere (wireless) Move from professional to consumer market shorter design cycle (months) higher integration level (price!) increasing need for design environments 3

4 Introduction Design environment Schematic capture Layout Layout extraction Output processor Yield optimisation Optimisation Statistical analysis Simulator Modelling tools Model library 4

5 Introduction RF designs/circuits are special they work at high frequencies (short wavelengths) Kirchhoff s laws not valid anymore (distributed and retardation effects). Maxwell is required. RF signal spectra are sparse noise (folding) and non-linearities are important (biterror rate) RF circuits require special simulation and modeling approaches 5

6 RF building blocks input/output blocks RF circuit simulation specifications: Noise, including noise folding Non-linearity 6

7 RF circuit simulation f out [2] [1] 2f LO [0] f LO [-1] [-2] V out = V LO x V in V in V out f out = f in ± f LO V LO -2f LO -f LO 0 f LO 2f LO f in -f LO -2f LO f out = f in + k*f PSS - y1-axis - DB(VN(PIF)).[-5] DB(VN(PIF)).[-4] DB(VN(PIF)).[-3] DB(VN(PIF)).[-2] DB(VN(PIF)).[-1] DB(VN(PIF)).[0] DB(VN(PIF)).[1] DB(VN(PIF)).[2] DB(VN(PIF)).[3] DB(VN(PIF)).[4] DB(VN(PIF)).[5] Ampl (db) G -4.0G -2.0G G 4.0G 6.0G Freq (Hz) 7

8 RF circuit simulation RF building blocks autonomous blocks - y1-axis - DBCSPP[0] Phase noise (db) specifications: k 159.0k k k k 159.1k k k f k Freq (Hz) Oscillation frequency Noise, especially phase noise 8

9 RF circuit simulation Common two-step approach to noise simulation: 1. Periodic Steady State (PSS) analysis 2. Linear or non-linear perturbation analysis 9

10 RF circuit simulation PSS is a generalization of DC (the solution after infinite time) Useful as a solution by itself (non-linearity) As a first step in perturbation analysis 10

11 RF circuit simulation (Accelerated) Poincaré Shooting: Newton process on F(x*)=x* Finite difference Harmonic balance Time domain methods Frequency domain methods 11

12 RF circuit simulation Perturbation (noise) analysis for input/output (forced) systems Assume 12

13 RF circuit simulation Perturbation (noise) analysis for autonomous systems Period of the solution is changed by noise If we assume x(t)=x PSS (t)+x n (t), x n (t) does not remain small. - = 13

14 RF circuit simulation Assume x(t)=x PSS (t+α(t))+x n (t) x n (t) the amplitude noise (remains small) α(t) describes the phase noise α(t) is determined from a scalar differential equation based on Floquet theory α(t) may be unbounded 14

15 Floquet theory RF Circuit Simulation 15

16 RF Circuit Simulation Solving for α(t) when n(t)=b(x PSS (t))b(t) All deterministic sources should be solved together (non-linear equation) 16

17 RF Circuit Simulation Solving for α(t) for non-deterministic signals. 17

18 EM simulation 18

19 Kirchhoff versus Maxwell EM simulation Circuit simulation is based on the Kirchhoff current (and voltage) laws and lumped elements elements connected by a zero impedance The zero impedance/lumped approach is an approximation The approximation becomes worse for higher frequencies (smaller wave lengths) 19

20 Kirchhoff versus Maxwell EM simulation With decreasing wave lengths the following effects become important Distributed effects (the circuit has size) Retardation effects (EM waves have finite speed) Skin effect (inhomogeneous current distributions) These effects are accurately described by Maxwell s equations 20

21 Maxwell s equations EM simulation 21

22 EM simulation Maxwell s equations (approximations) 1) DC conditions 22

23 EM simulation Maxwell s equations (approximations) 2) Quasi-static approach: Eddy current solution For no radiation losses 23

24 EM simulation Maxwell s equations (approximations) 3) Full-wave approach 24

25 EM simulation Requirements for IC EM simulation Frequency range 1-10GHz ε r =2 to 4, λ=15 to 20mm Die size 10 mm 2 Quasi static approach sufficient (?) Antennas on chip Full-wave required 25

26 EM simulation Solution methods for Maxwell s equations Finite Difference Method Finite Element Method Boundary Element Method Method of Moments Finite Difference Time Domain Frequency domain Time domain 26

27 EM simulation Maxwell and circuit simulation co-simulation of circuit simulator and EM simulator (very slow but sometimes required) Model generation EM circuit model lumped elements 27

28 EM simulation Maxwell and circuit simulation Model descriptions Y and S parameters (tables) frequency domain analysis time domain analysis is problematic (numerical convolution) turn into an equivalent network? model size 28

29 EM simulation Maxwell and circuit simulation Model descriptions equivalent circuit model (network of R,L and C) difficult to generate (from EM method or from S parameters) issues like stability of reduced/compressed models model size alternatives system of DAE G and C matrices 29

30 Anticipated directions EM simulation EM modelling approach is currently used most in PCB, hybrids and microwave designs Most successful are 2.5D Boundary Element methods (speed) For IC interconnect one could adjust the above methods include 3D effect (interconnect are walls ) include lossy substrate Further reduction of the models (ROM?) 30

31 RF circuit simulation Open Problems Phase noise in Phase-Locked Loops (PLLs) Multi-tone excitation More study into and verification of the phase noise perturbation approach?? How to deal with S-parameter models 31

32 EM modelling Open Problems RF IC interconnect modelling requires 3D effect Substrate must be taken into account This increases the complexity of the problem There is a need for fast model generator Model generators must be robust Current models are too large (complexity) 32