## Parametric yield maximization using gate sizing based on efficient statistical power and delay gradient computation (2005)

Venue: | In ICCAD |

Citations: | 21 - 2 self |

### BibTeX

@INPROCEEDINGS{Chopra05parametricyield,

author = {Kaviraj Chopra and Saumil Shah and Ashish Srivastava and David Blaauw and Dennis Sylvester},

title = {Parametric yield maximization using gate sizing based on efficient statistical power and delay gradient computation},

booktitle = {In ICCAD},

year = {2005},

pages = {1023--1028}

}

### OpenURL

### Abstract

With the increased significance of leakage power and performance variability, the yield of a design is becoming constrained both by power and performance limits, thereby significantly complicating circuit optimization. In this paper, we propose a new optimization method for yield optimization under simultaneous leakage power and performance limits. The optimization approach uses a novel leakage power and performance analysis that is statistical in nature and considers the correlation between leakage power and performance to enable accurate computation of circuit yield under power and delay limits. We then propose a new heuristic approach to incrementally compute the gradient of yield with respect to gate sizes in the circuit with high efficiency and accuracy. We then show how this gradient information can be effectively used by a non-linear optimizer to perform yield optimization. We consider both inter-die and intra-die variations with correlated and random components. The proposed approach is implemented and tested and we demonstrate up to 40 % yield improvement compared to a deterministically optimized circuit. 1.

### Citations

357 |
A neutral netlist of 10 combinational benchmark circuits and a target translator in fortran
- Brglez, Fujiwara
- 1985
(Show Context)
Citation Context ...erent power and timing constraints. Our results indicate that performing statistical optimization can significantly improve the timing yield of the design. We compare our results based on the ISCAS85 =-=[23]-=- benchmarks which were synthesized in a 130 nm technology. The yield optimization results are given in Table 2. The first sub-section including columns 2, 3, 4 and 5 report the initial timing and powe... |

215 |
Multivariate statistical methods
- Morrison
- 1976
(Show Context)
Citation Context ...arameter, which is defined using a correlation matrix. This gives a total of Ng RVs for each parameter, which are assumed to have a joint multinormal distribution. Using principal components analysis =-=[19]-=- the correlated component is expressed as a linear combination of Ng independent Gaussian RVs (zi), and the random variation in all the process parameters is lumped into a single RV - η for delay and ... |

201 | Statistical timing analysis considering spatial correlations using a single PERT-like traversal
- Chang, Sapatnekar
- 2003
(Show Context)
Citation Context ...p ⎛ Delay = d nom + ⎜α p i ⎝ ⎛ Leakage = exp ⎜ ⎜Vnom + ⎝ n ∑ ∑ = 1 i= 1 p ⎞ γ i zi ⎟ + η d R ⎠ n ⎛ ⎞ ⎞ ⎜ β z ⎟ + R ⎟ p γ i i ηl ⎝ i ⎠ ⎠ ∑ ∑ i= 1 = 1 Timing analysis is then performed in the spirit of =-=[3]-=-[17], and the delay is propagated through the circuit, while maintaining the node delays in the same canonical form with different coefficients. The sum operation is performed by simply adding the coe... |

138 |
The Greatest of a Finite Set of Random Variables
- Clark
- 1961
(Show Context)
Citation Context ...for the bivariate Gaussian distribution for c3540. of the summed pdfs. The max operation is performed by matching the mean, variance and the correlation of the max of two RV (which are obtained using =-=[20]-=-) and the canonical expression of the max. Leakage power analysis is based on summing lognormal RVs using Wilkinson’s method [21] as proposed in [8]. The leakage of each gate is iteratively added to t... |

106 |
LANCELOT: A Fortran Package for Large-Scale Nonlinear Optimization (Release A
- Toint
- 1992
(Show Context)
Citation Context ...t by integrating the perturbed bivariate Gaussian distribution over the region defined by the leakage power and timing constraint. This gradient computation technique is then integrated with LANCELOT =-=[18]-=-, a large-scale non-linear optimizer, to improve the parametric yield of the design, and is found to provide an 8X improvement in runtime with an average error of 0.1%. The remainder of the paper is o... |

90 |
On the sums of distribution function and moments of power sums with log-normal components. Bell Systems Technical Journal. Vol 61, pp 1441 to 1462. Mean error Standard Deviation BBC model 500 m resolution 5 dB 9.4 UKPM model 500 m resolution - 5 dB 8.3 UK
- Schwartz, Yeh
- 1982
(Show Context)
Citation Context ...nce and the correlation of the max of two RV (which are obtained using [20]) and the canonical expression of the max. Leakage power analysis is based on summing lognormal RVs using Wilkinson’s method =-=[21]-=- as proposed in [8]. The leakage of each gate is iteratively added to the sum which is maintained in canonical form. In each addition the coefficients of the canonical expression for the sum are calcu... |

41 | Fast statistical timing analysis handling arbitrary delay correlations - Orshansky, Bandyopadhyay - 2004 |

36 | Novel Sizing Algorithm for Yield Improvement under - Choi, Paul, et al. - 2004 |

29 | Statistical analysis of subthreshold leakage current for VLSI circuits
- Rao, Srivastava, et al.
- 2004
(Show Context)
Citation Context ...ion of the max of two RV (which are obtained using [20]) and the canonical expression of the max. Leakage power analysis is based on summing lognormal RVs using Wilkinson’s method [21] as proposed in =-=[8]-=-. The leakage of each gate is iteratively added to the sum which is maintained in canonical form. In each addition the coefficients of the canonical expression for the sum are calculated by matching t... |

28 | A methodology to improve timing yield in the presence of process variations. DAC - Raj, Vrudhula, et al. - 2004 |

22 | Statistical optimization of leakage power considering process variations using dual-vth and sizing - Srivastava, Sylvester, et al. |

7 |
et al., “Accurate and efficient gate-level parametric yield estimation considering correlated variations in leakage power and performance
- Srivastava
(Show Context)
Citation Context ...runtimes. Thus, there is a critical need to develop approaches that perform true and efficient parametric yield optimization, where yield is defined using both power and timing constraints. Recently, =-=[17]-=- proposed an approach to perform gate-level yield analysis in a computationally efficient manner while considering the correlation between power and performance. The approach was based on a principal-... |

6 | et al., “Full-Chip Sub-Threshold Leakage Power Prediction Model for - Narendra |

4 | Variability inspired implementation selection problem - Davoodi, Khandelwal, et al. - 2004 |

3 |
et al., “Parametric yield analysis and constrained-based supply voltage optimization
- Rao
- 2005
(Show Context)
Citation Context ...amples of a design are also expected to have higher power dissipation and vice-versa. This leads to a two-sided constraint on the feasible region of parametric yield defined by delay and power limits =-=[2]-=-, and causes significant yield loss under variations in process parameters. A number of analysis techniques to consider the impact of variability on timing [3-7] and power [8-9] have been developed. H... |

3 | et al., “Statistical timing analysis using bounds and selective enumeration - Agarwal - 2003 |

3 |
Uncertainty aware circuit optimization
- Bai
- 2002
(Show Context)
Citation Context ...lity on timing [3-7] and power [8-9] have been developed. However, these approaches do not estimate the true parametric yield of a design considering both power and performance correlation. Reference =-=[10]-=- was the first to consider the impact of variability on circuit optimization. The authors described the formation of a timing wall due to deterministic power optimization which increases the susceptib... |

2 |
et al., “Statistical timing based optimization using gate sizing
- Agarwal
- 2005
(Show Context)
Citation Context ...ents [17] a zi E( Pleake ) a zi ( P ) E( e ) ⎛ ⎞ ⎛ a ⎜ ⎟ = ⎜ i = log log ⎝ E leak ⎠ ⎝ ( ) ( ) ( ( ) ( ) ( ) ⎟⎟ b zi c zi E P + ⎞ leake E Pleake b c zi E P + E P E e Using the expressions developed in =-=[13]-=-, the remaining two P can be expressed as coefficients in the expression for a leak 1 ⎛ a = ⎜ 0 log 2 ⎜ ⎝ a ⎡ ( ( ) ( ) ( ( ) ( ) ( ) ( ) ( ) ⎟⎟ b b 4 E P + ⎞ leak E Pleak b b 2 E P + E P + Var P + Va... |

1 | et al., “First-order incremental block-based statistical timing analysis - Viswesweriah - 2004 |

1 | Block-Based statistical timing analysis with uncertainty - Devgan, Kashyap - 2003 |

1 | A new statistical algorithm for gate sizing - Mani, Orshansky - 2004 |

1 |
The bivariate normal integral,” Biometrika
- Cadwell
- 1951
(Show Context)
Citation Context ... and dynamic power of the design. The above yield expression is now equivalent to the integral of a bivariate Gaussian RV over a rectangular region, and can be evaluated using expression developed in =-=[22]-=-. Both the timing and power computation steps require O(nNg) steps, where n is the number of gates in the design and Ng is the number of regions into which the design is partitioned to capture the cor... |

1 |
Static timing analysis using backward propagation
- Lee, Zolotov, et al.
- 2004
(Show Context)
Citation Context ...cutset-source set (nodes within the dashed shape) for node 8 1 This neglects the impact of the change in the slope of the circuit. However, the approach for timing analysis using backward propagation =-=[24]-=- can be used to consider their impact within the same framework. 7 8 8 9 9 10 10sHowever, the AT for all edges that are in the fanout cone of the fanin-set of node n changes. However we are only inter... |