Advanced Tools for Simulation and Design of Oscillators/PLLs

13 downloads 0 Views 3MB Size Report
LTI. LPTV Our approach. Jitter from white noise. X. *. Flicker/coloured jitter. X ? *. Supply interference jitter X. *. VCO frequency control. *. Injection locking. X. X.
Advanced Tools for Simulation and Design of Oscillators/PLLs

Xiaolue Lai and Jaijeet Roychowdhury

University of Minnesota, Minneapolis, USA

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 1

Oscillators: A Special Simulation Challenge ●

Computation/size/accuracy: much greater than for amps/mixers ●

● ●



inefficient for even 1-transistor oscillators ● long startups: many cycles ● tiny timesteps needed per cycle integrated RF: 100s to 1000s of transistors errors dependent on size of timesteps, integration method, size of perturbations fundamental cause: marginal phase stability of all oscillators ● numerical errors integrate over time

 SPICE­based perturbation analysis of  oscillators is not a good idea ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 2

Alternatives to SPICE-level simulation ●

Use simplified phase macromodels ●



Obtain effects of external perturbations on phase directly without simulating the full circuit (Linear Time-Invariant (LTI) models) Linear VCO macromodel

Ken Kundert, Cadence

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 3

Limitations of Linear Phase Macromodels ●

Linear models are not generally applicable for all  effects LTI model: narrow applicability LPTV models, including:  Kaertner's LPTV IRF Hajimiri's closed­form ISF better, but validity depends on circuit/dynamical  effects  Eg, cannot capture nonlinear phenomena

● ●











injection locking, jitter, cycle slips, power grid noise effects, ...

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 4

Automatically-extracted Oscillator Macromodel works for ALL effects, ALL oscillators Injection locking

Coupled oscs

Phase noise, timing jitter

PLL design methodology

Power/ground interference jitter

Early design tools/formulae

PLL Lock/Capture PLL cycle slipping Automated Extraction Algorithms

ONE Small, Nonlinear

phase/amplitude macromodel (fast to simulate)

Nano/Bio applications Fast Simulation Algorithms

(Ring, LC, biological, ...)

Fast envelope methods

(SPICE-level circuit, or differential equations)

Oscillator AC

ANY oscillator

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 5

Automatically Extracted Nonlinear Oscillator Macromodel

Perturbation Projection Phase Vector (PPV/NISF) error/ jitter

Noise “Input”

Phase macromodel is nonlinear and scalar (nonlinear = captures complex dynamics) (scalar = small, fast to evaluate) Drop-in replacement for linear phase models ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 6

Feature Comparison with Existing Methods LTI

LPTV Our approach

Jitter from white noise Flicker/coloured jitter Supply interference jitter VCO frequency control Injection locking PLL lock/capture Cycle slipping Amplitude effects

   i  i i NA

Model generation speed Model gen. robustness Model simulation speed

Poor i Poor Excellent Poor Mediocre Good i Excellent Excellent Excellent

i  i i  i  NA

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

       

Slide 7

PLL Capture and Lock Transients

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 8

PLL Capture/Lock: fref=1.07f0 >1000x speedup over full simulation

Reference (full simulation) LTI Nonlinear ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 9

PLL Capture/Lock: fref = 1.074 f0 Nonlinear

Reference (full simulation)

LTI

>1000x speedup over full simulation

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 10

PLL Capture/Lock: fref = 1.083 f0

LTI

Nonlinear

Reference (full simulation)

>1000x speedup over full simulation

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 11

Oscillator and PLL jitter due to Supply Interference

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 12

Ring Oscillator: Per-cycle jitter as a function of sinusoidal supply interference frequency PPV, used in LPTV equation

Nonlinear

Reference (full simulation)

Closed-form (LPTV) ISF

~100x speedup over full simulation

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 13

PLL: Max jitter as function of supply interference frequency Closed-form (LPTV) ISF

Nonlinear

Reference (full simulation) >1000x speedup over full simulation ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 14

PLL Cycle Slipping

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 15

PLL: cycle slipping (5mA, 10T shock)

Nonlinear

LTI

Reference (full simulation)

>1000x speedup over full simulation ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 16

PLL: cycle slipping (3mA, 10T shock) Reference (full simulation)

Nonlinear

LTI >1000x speedup over full simulation ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 17

Impact of Nonlinear Macromodels on PLL Design Process

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 18

Injection-aided PLL structure HPF Reference frequency

PFD

switch

LPF LPF

VCO

1 N

Lock acquisition time reduction is important ● High-pass path: induces injection locking ●

speeds lock acquisition ● Once locked, HPF path is switched out of the loop ● to lower phase noise in lock ●

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 19

PLL Capture (freq shift vs time)

With HPF path Without HPF path



~3x faster lock acquisition due to injection aiding

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 20

Nonlinear PLL Macromodel: Benefits for Design ●

Linear macromodels: wrong results (though fast) ●



can mislead, waste design time

Nonlinear macromodels: can be trusted during design ●





sanity checking via SPICE-level simulation: still indispensable ● ~80 simulations in all ● ~5 full SPICE-level simulations ● macromodel offers 100-1000x speedup over full ● few minutes/simulation (including macromodel generation) Enables much more thorough exploration of design space ● many combinations of parameters, injection paths, ... ● Impractical using full simulation alone Design exploration completed in ~1 week ● Estimate: 10x saving in design time vs prior methodologies ● (with much more complete exploration)

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 21

Coupled Oscillator Systems

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 22

Brusselator biochemical oscillators Coupling (diffusion) terms Brusselator biochemical oscillators

“Network model” of coupled oscillators

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 23

Spontaneous Pattern Formation: 160,000 coupled Brusselator oscillators (g=1/15)

Simulation time: 200 minutes for 160 cycles (MATLAB); est speedup 1200x Slide 24

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Summary 

Nonlinear oscillator macromodel captures:     



injection locking PLL lock and capture PLL cycle slipping phase noise and jitter in oscillators and PLLs pattern formation in biochemistry

Productivity enhancer for in design flows 

10x faster?



Better, more completely explored designs

ASP-DAC 2007. Copyright © 2001-2006, J Roychowdhury and ASVG, Univ of Minnesota

Slide 25

Suggest Documents