Optimization method using parallel processors

• US 5,590,063 A
• Filed: 07/05/1994
• Issued: 12/31/1996
• Est. Priority Date: 07/05/1994
• Status: Expired due to Fees

• Current Assignee: Freescale Semiconductor, Inc. (NXP Semiconductors NV)
• Original Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Inventors: Miller, Monte G., Turner, Robert C., Halchin, David J., Golio, John M.
• Primary Examiner(s): Ramirez, Ellis B.
• Assistant Examiner(s): MILLER, CRAIG S
• Application Number:
  US08/270,948
• Time in Patent Office: 910 Days
• Field of Search: 364/580, 364/488, 364/164, 364/492, 364/402, 395/50, 395/24, 395/200, 395/325, 395/200.19
• US Class Current: 702/64
• CPC Class Codes: G06Q 10/04 Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" market predictions or forecasting for commercial activities G06Q30/0202

Abstract

A method executed by a computer for performing numerical optimization of arbitrary functions in a computer model using parallel processors (10, 12, 14). The method initializes (20) each processor with an initial estimate of the parameter value to be optimized. The initial estimate is evaluated (22) in each processor to determine a solution. A best estimate of the parameter value from the result in each processor is selected (24), and one or more of the parallel processors with the best estimate is set to run in gradient mode while the remaining processors run in random mode (26). The estimates of the parameter value from the processors running in random mode is evaluated until a local minimum is obtained from the processor running in gradient mode (28). The process is repeated until an optimal solution is found (34).

Claims

1. A computer implemented method of determining parameter values of a circuit model, comprising the steps of:

(a) initializing a plurality of processors with initial estimates of the parameter values of the circuit model;

(b) evaluating initial estimates within a plurality of processors respectively to determine a plurality of solutions;

(c) selecting a best estimate of the parameter values of the circuit model from said plurality of solutions;

(d) setting at least one of said plurality of processors having said best estimate to run in gradient mode while setting others of said plurality of processors to run in random mode; and

(e) evaluating estimates of the parameter values of the circuit model from said plurality of processors running in random mode until said at least one of said plurality of processors running in gradient mode determines a local minimum.

2. The method of claim 1 further including the step of repeating steps (c) through (e) until an optimal solution of the parameter values of the circuit model is determined as an absolute minimum value.

3. The method of claim 2 further including the step of initializing each of said plurality of processors with said initial estimates of the parameter values.

4. The method of claim 3 further including the step of storing said local minimum.

5. The method of claim 4 further including the step of setting said at least one of said plurality of processors running in gradient mode to run in random mode after finding said local minimum.

6. The method of claim 5 further including the step of taking a derivative of said estimate under evaluation by said at least one of said plurality of processors running in gradient mode to determine a slope of the parameter.

7. A computer implemented method of determining nodal voltage and currents that optimally describe electrical state of a circuit, comprising the steps of:

(a) initializing a plurality of processors with initial estimates of the nodal voltage and currents of the circuit;

(b) evaluating said initial estimates within each of said plurality of processors respectively to determine a plurality of solutions;

(c) selecting a best estimate of the nodal voltages and currents from said plurality of solutions;

(d) setting at least one of said plurality of processors having said best estimate to run in gradient mode while setting others of said plurality of processors to run in random mode; and

(e) evaluating estimates of the nodal voltages and currents from said plurality of processors running in random mode until said at least one of said plurality of processors running in gradient mode determines a local minimum.

8. The method of claim 7 further including the step of repeating steps (c) through (e) until an optimal solution of the nodal voltage and currents is determined as the absolute minimum value.

9. The method of claim 8 further including the step of storing said local minimum.

10. The method of claim 9 further including the step of setting said at least one of said plurality of processors running in gradient mode to run in random mode after finding said local minimum.

11. The method of claim 10 further including the step of taking a derivative of said estimate under evaluation by said at least one of said plurality of processors running in gradient mode to determine a slope of the nodal voltage and currents.

12. A computer implemented method of determining model parameter values that optimally describe a transistor circuit, comprising the steps of:

(a) initializing a plurality of processors with initial estimates of the model parameter values of the transistor circuit;

(b) evaluating initial estimates within a plurality of processors respectfully to determine a plurality of solutions;

(c) selecting a best estimate of the model parameter values of the transistor circuit from said plurality of solutions;

(d) setting at least one of said plurality of processors having said best estimate to run in gradient mode while setting others of said plurality of processors to run in random mode;

(e) evaluating estimates of the model parameter values of the transistor circuit from said plurality of processors running in random mode until said at least one of said plurality of processors running in gradient mode determines a local minimum; and

(f) repeating steps (c) through (e) until an optimal solution of the model parameter values of the transistor circuit is determined as an absolute minimum value.

13. The method of claim 12 further including the step of initializing each of said plurality of processors with said initial estimates of the model parameter values of the transistor circuit.

14. The method of claim 13 further including the step of storing said local minimum.

15. The method of claim 14 further including the step of setting said at least one of said plurality of processors running in gradient mode to run in random mode after finding said local minimum.

16. The method of claim 15 further including the step of taking a derivative of said estimate under evaluation by said at least one of said plurality of processors running in gradient mode to determine a slope of the model parameter.