US5396581A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Semiconductor neural network and operating method thereof
Assignee: Renesas Electronics Corp
•
Filing: 1993-01-19
•
Granted: 1995-03-07
•
Expires: 2013-01-19
Abstract
Abstract A semiconductor neural network includes a coupling matrix having coupling elements arranged in a matrix which couple with specific coupling strengths internal data input lines to internal data output lines. The internal data output lines are divided into groups. The neural network further comprises weighting addition circuits provided corresponding to the groups of the internal data cutput lines. A weighting addition circuit includes weighing elements for adding weights to signals on the internal data output lines in the corresponding group and outputting the weighted signals, and an addition circuit for outputting a total sum of the outputs of those weighting elements. The internal data output lines are arranged to form pairs and the addition circuit has a first input terminal for receiving one weighting element output of each of the pairs in common, a second input terminal for receiving the other weighting element output of each of the pairs in common, and a sense amplifier for differentially amplifying signals at the first and second input terminals. The neural network further includes a circuit for detecting a change time of an input signal, a circuit responsive to an input signal change for equalizing the first and second input terminals for a predetermined period, and a circuit for activating the sense amplifier after the equalization is completed. The information retention capability of each coupling element is set according to the weight of an associated weighting element. This neural network can provide multi-valued expression of coupling strength with less number of coupling elements.
Inventors
- 👤 Koichiro Mashiko
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-10 01:44
CPC Classifications
G; G06; G06N; G06N3/00; G06N3/02; G06N3/06; G06N3/063
Claims Analysis
Total Claims: 10
Claim 1
Independent Claim
233 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals together to produce an output signal, said step of adding including the steps of coupling all of said weighted signals to a first node for each of said groups to provide a summation signal, and amplifying said summation signal at said first node for output; detecting transition of incoming input data; and in response to said detected transition, establishing a predetermined first potential at said first node for a predetermined period.
Claim 2
Independent Claim
233 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals together to produce an output signal, said step of adding including the steps of coupling all of said weighted signals to a first node for each of said groups to provide a summation signal, and amplifying said summation signal at said first node for output; detecting transition of incoming input data; and in response to said detected transition, establishing a predetermined first potential at said first node for a predetermined period.
Claim 3
Independent Claim
307 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of group said plurality of internal output lines being arranged to form pairs in each group with each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising in each group of said plurality of internal output lines, after a step of coupling, with a specific coupling strength, an input line to a corresponding internal output line and transmitting signals between said input line and said corresponding output line, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals on said plurality of internal output lines together to produce an output signal, said step of adding for each group including the steps of transmitting said weighted signal from the other internal output line of each pair to a second node to provide a second summation signal, and differentially amplifying said first and second summation signals on said first and second nodes; detecting transition of incoming input data; in response to said detecting transition, equalizing potentials at each pair for a predetermined period; and in response to said detecting transition, enabling said differential amplification after said equalization is completed.
Claim 4
Independent Claim
307 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of group said plurality of internal output lines being arranged to form pairs in each group with each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising in each group of said plurality of internal output lines, after a step of coupling, with a specific coupling strength, an input line to a corresponding internal output line and transmitting signals between said input line and said corresponding output line, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals on said plurality of internal output lines together to produce an output signal, said step of adding for each group including the steps of transmitting said weighted signal from the other internal output line of each pair to a second node to provide a second summation signal, and differentially amplifying said first and second summation signals on said first and second nodes; detecting transition of incoming input data; in response to said detecting transition, equalizing potentials at each pair for a predetermined period; and in response to said detecting transition, enabling said differential amplification after said equalization is completed.
Claim 5
Independent Claim
297 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, said plurality of internal output lines being arranged to form pairs in each group with each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, after a step of coupling, with a specific coupling strength, an input line to a corresponding internal output line and transmitting signals between said input line and said corresponding output line, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals on said plurality of internal output lines together to produce an output signal, said step of adding for each group including the steps of transmitting said weighted signal from the other internal output line of each pair to a second node to provide a second summation signal, and differentially amplifying said first and second summation signals on said first and second nodes; detecting transition of incoming input data; and in response to said detecting transition, precharging each internal output line to a predetermined first potential for a predetermined period.
Claim 6
Independent Claim
297 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, said plurality of internal output lines being arranged to form pairs in each group with each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, after a step of coupling, with a specific coupling strength, an input line to a corresponding internal output line and transmitting signals between said input line and said corresponding output line, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals on said plurality of internal output lines together to produce an output signal, said step of adding for each group including the steps of transmitting said weighted signal from the other internal output line of each pair to a second node to provide a second summation signal, and differentially amplifying said first and second summation signals on said first and second nodes; detecting transition of incoming input data; and in response to said detecting transition, precharging each internal output line to a predetermined first potential for a predetermined period.
Claim 7
Independent Claim
309 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, said plurality of internal output lines being arranged to form pairs in each group each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, after a step of coupling, with a specific coupling strength, an input line to a corresponding internal output line and transmitting signals between said input line and said corresponding output line, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals on said plurality of internal output lines together to produce an output signal, said step of adding for each group including the steps of transmitting said weighted signal from the other internal output line of each pair to a second node to provide a second summation signal, and differentially amplifying said first and second summation signals on said first and second nodes; detecting transition of incoming input data; in response to said detecting transition, precharging said first and second nodes to a first potential for a predetermined period; and enabling said differentially amplifying after said precharging for said predetermined period is completed.
Claim 8
Independent Claim
309 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, said plurality of internal output lines being arranged to form pairs in each group each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, after a step of coupling, with a specific coupling strength, an input line to a corresponding internal output line and transmitting signals between said input line and said corresponding output line, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals on said plurality of internal output lines together to produce an output signal, said step of adding for each group including the steps of transmitting said weighted signal from the other internal output line of each pair to a second node to provide a second summation signal, and differentially amplifying said first and second summation signals on said first and second nodes; detecting transition of incoming input data; in response to said detecting transition, precharging said first and second nodes to a first potential for a predetermined period; and enabling said differentially amplifying after said precharging for said predetermined period is completed.
Claim 9
Independent Claim
226 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals together to produce an output signal, detecting transition of incoming input data; in response to an output signal of detecting means, generating a one-shot pulse signal having a predetermined short time interval; in response to said generated one-shot pulse signal, gating said incoming input data and transmitting the gated data to said plurality of internal input lines as said data signal to be processed.
Claim 10
Independent Claim
226 words
A method of weighting internal output lines of a semiconductor neural network including: a plurality of input lines each for transmitting a data signal to be processed; a plurality of internal output lines provided in a direction intersecting said plurality of input lines and divided into a plurality of groups, each group including a plurality of internal output lines and transmitting an internal output data signal; and a plurality of coupling elements provided at the intersections of said plurality of input lines and said plurality of internal output lines each for coupling, with a specific coupling strength, an input line with a corresponding internal output line and transmitting signals between said input line and said corresponding internal output line, said method comprising, in each group of said plurality of internal output lines, the steps of: loading a signal on each of said plurality of internal output lines with a previously selected weight to produce weighted signals; adding all of said weighted signals together to produce an output signal, detecting transition of incoming input data; in response to an output signal of detecting means, generating a one-shot pulse signal having a predetermined short time interval; in response to said generated one-shot pulse signal, gating said incoming input data and transmitting the gated data to said plurality of internal input lines as said data signal to be processed.
References Cited (Backward Citations)
US4660166A
Cited
US4982370A
Cited
US4901271A
Cited
US4807168A
Cited
US5004932A
Cited
US5021988A
Cited
US4988891A
Cited
US5053638A
Cited
US5146542A
Cited
US5101361A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1998-03-06 | 1998-09-05 | 1999-03-07 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2002-03-06 | 2002-09-05 | 2003-03-07 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2006-03-06 | 2006-09-05 | 2007-03-07 | $7700.00 | Window Closed / Lapsed |