## LARTTE: A Posynomial-Based Lagrangian Relaxation Tuning Tool for Fast and Effective Gate-Sizing and Multiple Vt Assignment

Citations: | 1 - 0 self |

### BibTeX

@MISC{Chou_lartte:a,

author = {Hsinwei Chou and Yu-hao Wang and Charlie Chung-ping Chen},

title = {LARTTE: A Posynomial-Based Lagrangian Relaxation Tuning Tool for Fast and Effective Gate-Sizing and Multiple Vt Assignment},

year = {}

}

### OpenURL

### Abstract

Abstract — In this paper, we propose a novel method for fast and effective gate-sizing and multiple Vt assignment using Lagrangian Relaxation (LR) and posynomial modeling. Our algorithm optimizes a circuit’s delay and power consumption subject to slew rate constraints, and can readily take process variation into account. We first use SPICE to generate accurate delay and power models in posynomial form for standard cells, then formulate a large-scale, convex optimization problem based on these models. Finally, we perform LR to solve for the globally-optimal 1 set of transistor sizes and Vts (with discretization) for each gate. Our key contribution is that we show for the first time that using posynomial models, LRbased circuit tuning can be carried out in a ”generalized ” or non-Gauss-Seidel manner for improved accuracy. Experimental results show that our implemented tuning tool, LARTTE, exhibits linear runtime and memory usage requirement, can effectively tune a circuit with over 15,000 variables and 8,000 constraints in under 7 minutes, and can minimize the probability of final delay variation by introducing a margin of separation between the worst output arrival time and all other outputs ’ arrival times. I.

### Citations

1931 |
Numerical Optimization
- Nocedal, Wright
- 2006
(Show Context)
Citation Context ...ary node Fig. 1. A combinational circuit. Gates NG = 1 3 2 Primary outputs PO = 2 1 0 0: auxiliary node B. Background: Optimization Problems and Posynomial Functions In general, optimization problems =-=[12]-=- have the form: minimize f0(x) subject to gi(x) ≤ 0, i=1,...,m hi(x) =0,i=1,...,n (1) where x ∈ℜ n is a n-vector of optimization variables and f0, gi, and hi are the objective function, inequality con... |

336 | SNOPT: an SQP algorithm for large-scale constrained optimization[J
- Gill, Murray, et al.
- 1997
(Show Context)
Citation Context ...final delay variation by introducing a margin of separation between the worst output arrival time and all other outputs’ arrival times. Experiments also show that LARTTE compares favorably with SNOPT =-=[11]-=-, a state-ofthe-art general-purpose optimization problem solver. LARTTE is over 250x faster than SNOPT, but can achieve the same quality of results. This paper is organized as follows. Background and ... |

271 | A limited memory algorithm for bound constrained optimization
- Byrd, Lu, et al.
- 1995
(Show Context)
Citation Context ...accuracy, efficiency, and elegance of our generalized solving-procedure is the key to LARTTE’s performance, and is the main contribution of this work. We resort to an off-the-shelf solver in L-BFGS-B =-=[15]-=- to solve LRS/λ. L-BFGS-B implements the well-known, LimitedMemory BFGS method [12], which has been proven to be exceptional for handling large-scale, unconstrained problems. This method belongs to th... |

97 |
Modeling and analysis of manufacturing variations
- Nassif
- 2001
(Show Context)
Citation Context ...e power reduction must be addressed in the circuit tuning process. As CMOS scales into nanometer technology, another problem that must be dealt with in the tuning process is on-chip process variation =-=[6]-=-. From device geometry to device parameters, many things can vary to cause the final critical delay value to differ non-trivially from that calculated via static timing analysis. This in turn makes yi... |

81 | Fast and exact simultaneous gate and wire sizing by lagrangian relaxation
- Chen, Chu, et al.
- 1998
(Show Context)
Citation Context ...blem size/complexity explosion, the task of circuit or transistorlevel tuning (for delay, power, noise, etc.) can be extremely timeconsuming and even overwhelming for today’s designers. In past works =-=[1]-=- [2], it was shown that the method of Lagrangian Relaxation (LR) can be used to solve the circuit tuning problem efficiently and effectively. However, there are several flaws with these earlier works,... |

52 | Global convergence of a class of trust region algorithms for optimization with simple bounds
- Conn, Gould, et al.
- 1988
(Show Context)
Citation Context ...dient approach shown in Table II, line 3, to form a new guess at every iteration. θk is a step size value which is initialized to 1 and gradually updated over iterations using a Trust-Region approach =-=[14]-=-. We continue to iterate and make new guesses for the correct, optimal value of λ until our LRS/λ ∗ value converges to that of the PP value. When this occurs, we will have found our desired multiplier... |

36 | Design and Optimization of Low Voltage High Performance Dual Threshold
- Wei
- 1998
(Show Context)
Citation Context ...imple-yet-inaccurate Elmore delay model. Furthermore, as the magnitude of leakage power is quickly catching up to that of dynamic power [3], the important task of using multiple Vt levels in a design =-=[4]-=- [5] for leakage power reduction must be addressed in the circuit tuning process. As CMOS scales into nanometer technology, another problem that must be dealt with in the tuning process is on-chip pro... |

25 |
A New Class of Convex Functions for Delay Modeling and their Application to the Transistor Sizing Problem
- Kasamsetty, Ketkar, et al.
(Show Context)
Citation Context ...’s delay and power consumption subject to slew rate constraints, and can readily take process variation into account. We first use SPICE to generate accurate delay and power models in posynomial form =-=[10]-=- for standard cells, then formulate a large-scale, convex optimization problem based on these models. Finally, we perform LR to solve for the globally-optimal (with respect to the posynomial-based opt... |

19 |
C.: Gate sizing using Lagrangian relaxation combined with a fast gradient-based pre-processing step
- Tennakoon, Sechen
(Show Context)
Citation Context ... size/complexity explosion, the task of circuit or transistorlevel tuning (for delay, power, noise, etc.) can be extremely timeconsuming and even overwhelming for today’s designers. In past works [1] =-=[2]-=-, it was shown that the method of Lagrangian Relaxation (LR) can be used to solve the circuit tuning problem efficiently and effectively. However, there are several flaws with these earlier works, suc... |

19 | Uncertainty-aware circuit optimization
- Bai, Visweswariah, et al.
- 2002
(Show Context)
Citation Context ...t discretization. size tuning, since a well-tuned circuit that neglects process variation during the tuning process typically exhibits a ”wall-like” distribution in its primary outputs’ arrival times =-=[7]-=-. This is because to squeeze the most performance benefits out of sizing, a standard tuner will size in such a way that most, if not all, of the primary outputs’(PO) arrival times end up with equally ... |

12 | Tits, “User’s guide for cfsqp version 2.4: A C code for solving (large scale) constrained nonlinear (minimax) optimization problems, generating iterates satisfying all inequality constraints,” Institute for Systems Research - Lawrence, Zhou, et al. - 1996 |

8 | Design methodology and optimization strategy for dual-V TH scheme using commercially available tools,” ISLPED
- Hirabayashi, Nose, et al.
- 2001
(Show Context)
Citation Context ...e-yet-inaccurate Elmore delay model. Furthermore, as the magnitude of leakage power is quickly catching up to that of dynamic power [3], the important task of using multiple Vt levels in a design [4] =-=[5]-=- for leakage power reduction must be addressed in the circuit tuning process. As CMOS scales into nanometer technology, another problem that must be dealt with in the tuning process is on-chip process... |

1 | Increase in delay uncertainty by delay optimization - Hashimoto, Onodera |