## Minimum Dynamic Power CMOS Circuit Design by a Reduced Constraint Set Linear Program (2003)

Venue: | in Proc. of 16th International Conference on VLSI Design |

Citations: | 17 - 10 self |

### BibTeX

@INPROCEEDINGS{Raja03minimumdynamic,

author = {Tezaswi Raja and Vishwani D. Agrawal and Michael L. Bushnell},

title = {Minimum Dynamic Power CMOS Circuit Design by a Reduced Constraint Set Linear Program},

booktitle = {in Proc. of 16th International Conference on VLSI Design},

year = {2003},

pages = {527--532}

}

### Years of Citing Articles

### OpenURL

### Abstract

In the previous work, the problem of nding gate delays to eliminate glitches has been solved by linear programs (LP) requiring an exponentially large number of constraints. By introducing two additional variables per gate, namely, the fastest and the slowest arrival times, besides the gate delay,we reduce the number of the LP constraints to be linear in circuit size. For example, the 469-gate c880 circuit requires 3,611 constraints as compared to the 6.95 million constraints needed with the previous method. The reduced constraints provably produce the same exact LP solution as obtained by the exponential set of constraints. For the rst time, we are able to optimize all ISCAS'85 benchmarks. For the c7552 circuit, when the input to output delay is constrained not to increase, a design with 366 delay bu ers consumes only 34 % peak and 38 % average power as compared to an unoptimized design. As shown in previous work, the use of delay bu ers is essential in this case. The practicality of the design is demonstrated by implementing an optimized 4-bit ALU circuit for which the power consumption was obtained by a circuit-level simulator. 1.

### Citations

391 |
AMPL: A Modeling Language for
- Fourer, Gay, et al.
(Show Context)
Citation Context ...5 10 4 10 3 6.95M 3611 c880 New model Old model 5. Results 5.1. Experimental procedure We use a C++ program to parse the logic level circuit netlist and generate the constraint set in the AMPL format =-=[11]-=-. This constraint set is read into AMPL and the optimized delays of gates and bu ers are obtained. We then use a power estimator [19] to compute the power savings obtained with new delays for the gate... |

236 | A Survey of Power Estimation Techniques in VLSI Circuits (Invited Paper
- Najm
- 1994
(Show Context)
Citation Context ...er of bu ers inserted for that maxdelay. Thepower estimation was done with a variable delay event-driven simulator that counts the number of all gate transitions produced by stuck-at fault vector set =-=[14, 17,13, 19]-=-. The reference case (shown as unoptimized) is a unit delay circuit with no bu ers added. The second row gives the result for the optimized circuit when the I/O delay is allowed to increase, sometimes... |

149 |
Low Power Design Methodology
- Rabaey, Pedram
- 1996
(Show Context)
Citation Context ...r, in reality they produce many transitions. These extra transitions are caused by the di erential delays of paths leading to the inputs of the gates. This subject is widely discussed in recent books =-=[7, 9, 16, 18, 20]-=-. Visiting Professor, Dept. of ECE, Rutgers University, Piscataway, NJ Among various methods for minimizing the dynamic power is the balanced delay method in which we equalize the delays of all paths ... |

139 |
DAGON:Technology Binding and Local Optimization by DAG
- Keutzer
- 1987
(Show Context)
Citation Context ...me elapsed (nanoseconds) Figure 4: Energy consumption of optimized and unoptimized 4-bit ALU circuits obtained by circuit-level simulation. cient techniques of technology mapping that can be employed =-=[15, 25]-=-. We are continuing this experiment to produce a layout and incorporate the routing delays in the constraint set. 7. Conclusion The main result of this work is a new formulation of theLPwithnumber of ... |

74 | Technology decomposition and mapping targeting low power dissipation
- Tsui, Pedram, et al.
- 1993
(Show Context)
Citation Context ...me elapsed (nanoseconds) Figure 4: Energy consumption of optimized and unoptimized 4-bit ALU circuits obtained by circuit-level simulation. cient techniques of technology mapping that can be employed =-=[15, 25]-=-. We are continuing this experiment to produce a layout and incorporate the routing delays in the constraint set. 7. Conclusion The main result of this work is a new formulation of theLPwithnumber of ... |

59 |
Accurate Simulation of Power Dissipation in VLSI Circuits
- Kang
- 1986
(Show Context)
Citation Context ...er of bu ers inserted for that maxdelay. Thepower estimation was done with a variable delay event-driven simulator that counts the number of all gate transitions produced by stuck-at fault vector set =-=[14, 17,13, 19]-=-. The reference case (shown as unoptimized) is a unit delay circuit with no bu ers added. The second row gives the result for the optimized circuit when the I/O delay is allowed to increase, sometimes... |

40 |
Optimization-Based Transistor Sizing
- Shyu, Sangiovanni-Vincentelli, et al.
- 1988
(Show Context)
Citation Context ...delay of the transistor. This method does not add any bu ers but solves the problem in a large dimensional space by treating all transistors as parameters and numerical convergence is often a problem =-=[23]-=-. Another method is called gate sizing where each logic gate is modeled as an equivalent inverter [3, 4]. The gate sizes varyinacontinuous manner. The main problems are the non-linearity of the model ... |

35 |
Low-Power CMOS VLSI Circuit Design
- Roy, Prasad
- 2000
(Show Context)
Citation Context ...r, in reality they produce many transitions. These extra transitions are caused by the di erential delays of paths leading to the inputs of the gates. This subject is widely discussed in recent books =-=[7, 9, 16, 18, 20]-=-. Visiting Professor, Dept. of ECE, Rutgers University, Piscataway, NJ Among various methods for minimizing the dynamic power is the balanced delay method in which we equalize the delays of all paths ... |

16 |
Computer-Aided Design Techniques for Low Power Sequential Logic Circuits
- Monteiro, Devadas
- 1996
(Show Context)
Citation Context ...r, in reality they produce many transitions. These extra transitions are caused by the di erential delays of paths leading to the inputs of the gates. This subject is widely discussed in recent books =-=[7, 9, 16, 18, 20]-=-. Visiting Professor, Dept. of ECE, Rutgers University, Piscataway, NJ Among various methods for minimizing the dynamic power is the balanced delay method in which we equalize the delays of all paths ... |

13 |
Low Power Design by Hazard Filtering
- Agrawal
- 1997
(Show Context)
Citation Context ...u ers on selected fanout branches. While bu ers consume power they allow the balancing without increase in overall delay of the circuit. An alternative to the balanced delay methodisthehazard ltering =-=[1]-=-. If a pulse of width lesser than the inertial delay of the gate is incident on a gate input then that would be suppressed or ltered by the gate and this is known as the ltering e ect of a gate [26]. ... |

13 |
Parhi, “Fast and exact transistor sizing based on iterative relaxation
- Sundararajan, Sapatnekar, et al.
- 2002
(Show Context)
Citation Context ...gn parameter. Once again the number of these constraints increases exponentially. This high complexity limits the model from optimizing large circuits. Another technique is known as transistor sizing =-=[5, 8, 10, 27, 21, 6, 24]-=-. By sizing, we mean the width and length of the transistor that change the driving capacity and thus the delay of the transistor. This method does not add any bu ers but solves the problem in a large... |

12 | Using Gate Sizing to Reduce Glitch Power
- Berkelaar, Jacobs
- 1996
(Show Context)
Citation Context ...onal space by treating all transistors as parameters and numerical convergence is often a problem [23]. Another method is called gate sizing where each logic gate is modeled as an equivalent inverter =-=[3, 4]-=-. The gate sizes varyinacontinuous manner. The main problems are the non-linearity of the model and the discrete gate sizes which cause mathematical di culties especially for large circuits. In this p... |

12 |
Nossek, “Automated transistor sizing algorithm for minimizing spurious switching activities
- Wróblewski, Schimpfle, et al.
(Show Context)
Citation Context ...gn parameter. Once again the number of these constraints increases exponentially. This high complexity limits the model from optimizing large circuits. Another technique is known as transistor sizing =-=[5, 8, 10, 27, 21, 6, 24]-=-. By sizing, we mean the width and length of the transistor that change the driving capacity and thus the delay of the transistor. This method does not add any bu ers but solves the problem in a large... |

9 |
ASAP: A Transistor Sizing Tool for Area, Delay and Power Optimization of CMOS Circuits
- Datta, Nag, et al.
- 1994
(Show Context)
Citation Context ...gn parameter. Once again the number of these constraints increases exponentially. This high complexity limits the model from optimizing large circuits. Another technique is known as transistor sizing =-=[5, 8, 10, 27, 21, 6, 24]-=-. By sizing, we mean the width and length of the transistor that change the driving capacity and thus the delay of the transistor. This method does not add any bu ers but solves the problem in a large... |

8 | Computing Entire Area/Power Consumption versus Delay Trade-o Curve for Gate Sizing Using a Piecewise Linear Simulator - Berkelaar, Buurman, et al. - 1996 |

8 |
Minimizing Power Consumption of Static CMOS Circuits by Transistor Sizing and Input Reordering
- Borah, Irwin, et al.
- 1995
(Show Context)
Citation Context |

7 |
A Reduced Constraint Set Linear Program for Low Power Design of Digital Circuits
- Raja
- 2002
(Show Context)
Citation Context ...he discrete gate sizes which cause mathematical di culties especially for large circuits. In this paper, we describe a new technique in which the constraint set size of the LP has a linear complexity =-=[19]-=-. 2. The New Approach In addition to the single delay parameter as was done earlier [1,2, 4], we introduce two newvariables for every gate, one for earliest time and the other for the most delayed tim... |

6 |
Digital Circuit Design for Mimimum Transient Energy and Linear Programming Method
- Agrawal, Bushnell, et al.
- 1999
(Show Context)
Citation Context ...iving inputs at the gate, the glitches can be eliminated. Clearly, the overall delay constraint will increase. Acombination of both delay balancing and hazard ltering has been tried by Agrawal et al. =-=[2]-=-. They describe a linear programming model to generate constraints for hazard ltering while keeping the overall delay within limits. Consider a gate with two inputs 1 and 2. The minimum transient ener... |

6 |
Transistor Sizing in MOS Digital Circuits with Linear Programming
- Berkelaar, Jess
- 1990
(Show Context)
Citation Context |

3 |
TILOS: a Posynomial Programming Approach toTransistor Sizing
- Fishburn, Dunlop
- 1985
(Show Context)
Citation Context |

3 |
Timing Veri cation and the Timing Analysis Program
- Sr
- 1982
(Show Context)
Citation Context ...ans that events always occur in the interval [ti,Ti] at the output of the gate i. This technique of minimum and maximum arrival time variables is similar to the non-enumerative static timing analysis =-=[12]-=-. Theorem 1: Consider a gate i with n inputs, receiving events from fanin gates 1� 2�:::� n at times m1, m2, :::mn. Assuming that T1 T2 ::: Tn and t1 t2 ::: tn the number of events at the output of ga... |

3 |
Asynchronous Switching Circuits
- Unger
- 1969
(Show Context)
Citation Context ...ng [1]. If a pulse of width lesser than the inertial delay of the gate is incident on a gate input then that would be suppressed or ltered by the gate and this is known as the ltering e ect of a gate =-=[26]-=-. Thus, by adjusting the inertial delay to be greater than the di erential path delay of arriving inputs at the gate, the glitches can be eliminated. Clearly, the overall delay constraint will increas... |

2 |
E ects of Delay ModelinPeak Power Estimation of VLSI Circuits
- Hsiao, Rudnick, et al.
- 1997
(Show Context)
Citation Context ...er of bu ers inserted for that maxdelay. Thepower estimation was done with a variable delay event-driven simulator that counts the number of all gate transitions produced by stuck-at fault vector set =-=[14, 17,13, 19]-=-. The reference case (shown as unoptimized) is a unit delay circuit with no bu ers added. The second row gives the result for the optimized circuit when the I/O delay is allowed to increase, sometimes... |

2 |
Nassek, \Transistor Sizing for Switching Activity Reduction in Digital Circuits
- e, Wroblewski, et al.
- 1999
(Show Context)
Citation Context |

1 |
CMOS Digital Circuit Technology. Upper Saddle River
- Shoji
- 1988
(Show Context)
Citation Context ...pacitor dropped gradually as the circuit consumed energy. Thisvoltage drop, though not signi cant enough to change the operation of the circuit, gave an estimate of the energy consumed by the circuit =-=[22]-=-. The circuit was simulated on Spectre (by Cadence) for 50 random vectors and the energy consumed by the capacitor calibrated in picojoules is shown by the dotted curve in Figure 4. The gate delays de... |