![Example: Gainboosted op. amplifier
* [M.Waltari, “circuit techniques for low-voltage and high-speed A/D converters, PhD.]
Jiri Ocenasek, Magwel NV, Belgium 5](https://image.slidesharecdn.com/ocenasek2008obupm-1219681911556501-8/85/Robust-Circuit-Sizing-EDA-for-EDA-5-320.jpg)
Ad
More Related Content
Viewers also liked (10)
Similar to Robust-Circuit-Sizing:-EDA-for-EDA (20)
Ad
Recently uploaded (20)
Ad
Robust-Circuit-Sizing:-EDA-for-EDA
- 1. + July 2008 Jiri Ocenasek, Magwel NV, Belgium 1
- 2. Robust circuit sizing: EDA for EDA Jiri Ocenasek, Magwel NV, Belgium 2
- 3. Company Overview MAGWEL was founded in April 2003 Spinoff from IMEC in Belgium Initially focused on 3D deviceEM simulation With acquisition of Kimotion Technologies − Expanding into Robust Design and Verification of Analog and RF circuits Customers: TI, NXP, IMEC, UCL …. Participant in EU research projects Jiri Ocenasek, Magwel NV, Belgium 3
- 4. Robust sizing of Analog & RF building blocks Design variables Specifications Process and Environment EldoTM SpectreTM HspiceTM KtModels ADVanceMSTM … Performance models Optimized netlist Specification corners Jiri Ocenasek, Magwel NV, Belgium 4
- 5. Example: Gainboosted op. amplifier * [M.Waltari, “circuit techniques for low-voltage and high-speed A/D converters, PhD.] Jiri Ocenasek, Magwel NV, Belgium 5
- 6. Terminology I Design variables D − The designer can prescribe them within some accuracy − Example: W and L of transistors, bias current, … Operational variables O − The circuit has to work under all operational conditions within specified limits − Example: Temperature, Supply voltage, … Technology variables T − Variation of manufacturing process, statistically distributed − Example: Doping parameters of substrate Jiri Ocenasek, Magwel NV, Belgium 6
- 7. Terminology II Mismatch variables M Mismatch in T + Mismatch in D IDS1 IDS2 Transistors are not identical ! VGS1 VGS2 Jiri Ocenasek, Magwel NV, Belgium 7
- 8. Terminology III Response variables R − Extracted from simulator’s output: r = simulate(d,o,t,m ) − Example: Noise, Gain, Power, Die area, ... Constraint cost for ith constraint Ci(r) Ci(r) ≥ 0 if satisfied, Ci(r) < 0 if violated Example: Power is below 100 mW: “100mW Power” Goal cost for jth goal Gj(r) Jiri Ocenasek, Magwel NV, Belgium 8
- 9. Yield explanation Probability density function ρ(t,m ) Typically normal or uniform Advanced manufacturing processes the design d has to be robust against variations of t and m yield(d) = ∫t,m F(d,t,m) ρ(t,m) dt dm where F(d,t,m) indicates whether all constraints Ci are satisfied under all permissible operational conditions o: F(d,t,m)=1 if ∀i, ∀ο ϵΟ : Ci(simulate(d,o,t,m))≥0 F(d,t,m)=0 otherwise Jiri Ocenasek, Magwel NV, Belgium 9
- 10. Optimization target First stage: nominal optimization − find d such that ∀i: Ci(simulate(d,o=nominal,t=0,m=0))≥0 − can be replaced by userprovided solution Second stage: yield maximization − find/improve d such that yield(d) is maximized Third stage: when yield( d ) > 0.997 − minimize sum of goals ∑j Gj(d,o=nominal,t=0,m=0,r) while keeping the yield Jiri Ocenasek, Magwel NV, Belgium 10
- 11. Results: gainboosted amplifier # devices 56 # design variables 22 # constraints 63 # tech.+mm. variables 30 # operational variables 4 starting point none # simulations 12626 runtime 251 min total yield 99.8% goal 1 area = 7970 μm2 goal 2 power = 13.7 mW Jiri Ocenasek, Magwel NV, Belgium 11
- 12. Results: gainboosted amplifier Number of fitness evaluations Jiri Ocenasek, Magwel NV, Belgium 12
- 14. Why are Estimation of Distribution Algorithms suitable? Decomposability of the problem − the circuit is composed of subblocks Complicated fitness landscape − multimodality: several local optima − nonlinearities: responses are hard to be modeled − discontinuities: simulator switches modes − missing data: regions where simulation fails Jiri Ocenasek, Magwel NV, Belgium 14
- 15. Conclusion Robust Optimizer − Can handle tough circuits where others fail − Capable of handling fairly large blocks Produces quality results with high yield − 3 step approach, does not require user intervention − Provides insight: yield distributions, sensitivities, tradeoffs Integration − Cadence design environment − Simulators: Spectre, Eldo, Hspice (others can be added) − Supports LSF for distributed simulation Jiri Ocenasek, Magwel NV, Belgium 15