US5146602A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method of increasing the accuracy of an analog neural network and the like
Assignee: Intel Corp
•
Filing: 1990-12-26
•
Granted: 1992-09-08
•
Expires: 2010-12-26
Abstract
Abstract A method for increasing the accuracy of an analog neural network which computers a sum-of-products between an input vector and a stored weight pattern is described. In one embodiment of the present invention, the method comprises initially training the network by programming the synapses with a certain weight pattern. The training may be carried out using any standard learning algorithm. Preferably, a back-propagation learning algorithm is employed. Next, network is baked at an elevated temperature to effectuate a change in the weight pattern previously programmed during initial training. This change results from a charge redistribution which occurs within each of the synapses of the network. After baking, the network is then retrained to compensate for the change resulting from the charge redistribution. The baking and retraining steps may be successively repeated to increase the accuracy of the neural network to any desired level.
Inventors
- 👤 Mark A. Holler
- 👤 Simon M. Tam
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-30 17:06
CPC Classifications
H; H10; H10D; H10D30/00; H10D30/60; H10D30/68; H10D30/687; G; G06; G06N; G06N3/00; G06N3/02; G06N3/06; G06N3/063; G06N3/065; G11; G11C; G11C27/00; G11C27/005
Claims Analysis
Total Claims: 13
Claim 1
Independent Claim
82 words
In a neural network having a plurality of programmable synapses for computing a sum-of-products between an input vector and a stored weight vector, a method of increasing the accuracy of said neural network comprising the steps of: (a) initially training said neural network by programming said synapses with a certain weight vector; (b) baking said neural network to effectuate a change in said certain weight vector; and (c) retraining said neural network to compensate for said change resulting from said baking step.
Claim 2
Dependent Claim
82 words
In a neural network having a plurality of programmable synapses for computing a sum-of-products between an input vector and a stored weight vector, a method of increasing the accuracy of said neural network comprising the steps of: (a) initially training said neural network by programming said synapses with a certain weight vector; (b) baking said neural network to effectuate a change in said certain weight vector; and (c) retraining said neural network to compensate for said change resulting from said baking step.
Claim 3
Dependent Claim
25 words
The method of claim 1 further comprising the step of: (d) repeating steps (b) and (c) to further increase the accuracy of said neural network.
Claim 4
Dependent Claim
17 words
The method of claim 1 wherein said training and retraining steps each employ a back-propagation learning algorithm.
Claim 5
Dependent Claim
29 words
The method of claim 3 wherein said plurality of said synapses each comprise a floating gate memory device which is programmed by transferring charge onto a floating gate member.
Claim 6
Independent Claim
145 words
A method of increasing the accuracy of a neural network which employs floating gate devices to store a weight vector, each of said floating gate devices comprising a cell having a control gate insulated from a floating gate by a dielectric layer, said dielectric layer being subject to charge redistribution which causes a change in the electrical state of said floating gate over time, said method comprising the steps of: (a) setting target weight values on said floating gate devices by programming each said cell; (b) placing said neural network in a high temperature environment for a predetermined amount of time in order to accelerate charge redistribution in said dielectric layer; (c) removing said neural network from said high temperature environment; and (d) retraining said neural network to correct for changes in said target weight values which have occurred as a result of step (b).
Claim 7
Independent Claim
145 words
A method of increasing the accuracy of a neural network which employs floating gate devices to store a weight vector, each of said floating gate devices comprising a cell having a control gate insulated from a floating gate by a dielectric layer, said dielectric layer being subject to charge redistribution which causes a change in the electrical state of said floating gate over time, said method comprising the steps of: (a) setting target weight values on said floating gate devices by programming each said cell; (b) placing said neural network in a high temperature environment for a predetermined amount of time in order to accelerate charge redistribution in said dielectric layer; (c) removing said neural network from said high temperature environment; and (d) retraining said neural network to correct for changes in said target weight values which have occurred as a result of step (b).
Claim 8
Dependent Claim
26 words
The method of claim 5 further comprising the step of: (e) repeating steps (b), (c) and (d) to further increase the accuracy of said neural network.
Claim 9
Dependent Claim
19 words
The method of claim 5 wherein said high temperature environment exceeds the normal operating temperature of said neural network.
Claim 10
Dependent Claim
14 words
The method of claim 7 wherein said high temperature environment is approximately 250° C.
Claim 11
Dependent Claim
19 words
The method of claim 5 wherein said dielectric comprises a stack of silicon dioxide, silicon nitride and silicon dioxide.
Claim 12
Dependent Claim
14 words
The method of claim 9 wherein said retraining step employs a back-propagation learning algorithm.
Claim 13
Dependent Claim
22 words
The method of claim 10 wherein said high temperature environment is approximately 250° C. and said predetermined time is about 24 hours.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1995-09-08 | 1996-03-09 | 1996-09-08 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1999-09-08 | 2000-03-09 | 2000-09-08 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2003-09-08 | 2004-03-09 | 2004-09-08 | $7700.00 | Window Closed / Lapsed |