US5086405A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Floating point adder circuit using neural network
Assignee: Samsung Electronics Co Ltd
•
Filing: 1990-07-10
•
Granted: 1992-02-04
•
Expires: 2010-07-10
Abstract
Abstract A floating point adder circuit using neural network concepts and having high speed operation is obtained by a controlling circuit using a comparator and an operating circuit using an adder and a subtractor.
Inventors
- 👤 Ho-sun Chung
- 👤 Seung-yeob Paek
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-04 07:52
CPC Classifications
G; G06; G06F; G06F17/00; G06F17/10; G06F7/00; G06F7/38; G06F7/48; G06F7/483; G06F7/485; G06N; G06N3/00; G06N3/02; G06N3/06; G06N3/063; G06F2207/00; G06F2207/38; G06F2207/48; G06F2207/4802
Claims Analysis
Total Claims: 8
Claim 1
Independent Claim
406 words
A floating point adder circuit using neural network which adds two binary numbers of K(M+N) bit that consist of an M-bit exponent part and an N-bit mantissa part comprising; a comparator for comparing the exponent magnitude of said two binary numbers by comprising first input synapses for connecting a first power supply voltage to a line as the connecting strength of the weight of each bit according to a first exponent value of said M-bit; second input synapses for connecting a second power supply voltage to said line as the connecting strength of the weight of each bit according to a second exponent value of said M-bit; and a neuron which is stimulated or restrained according to the difference of the connecting strength between said first power supply voltage and said second power supply voltage that are connected to said line; an adder which add the mantissa parts of said two binary numbers by comprising input synapses for respectively connecting said second power supply voltage to a first line of N+1 numbers as the connecting strength of the weight of each bit according to a first mantissa value of said N-bit, a second mantissa value of said N-bit and a carry input of one-bit; first bias synapses for connecting said first power supply voltage to each first line as the connecting strength of a multiple of said input connecting strength sum; feedback synapses for connecting said second power supply voltage to a second line of N+1 numbers of each lower output bit as the connecting strength of the weight of upper output bit; second bias synapses for connecting said first power supply voltage to the least significant bit line among said second lines as the connecting strength which subtracts 0.5 from the connecting strength of said first bias synapse and to the other lines as the connecting strength which subtracts the weight of the bit from the connecting strength of each lower bit line, in order to bias said each second line connected to said feedback synapses; neurons which are stimulated or restrained by comparing the difference of the connecting strength of said first power supply voltage and said second power supply voltage which are respectively connected to said first line and said second line; and inverters for inverting the output of each of said neurons; and a subtractor comprising said adder and inverters for inverting some mantissa value of said N-bit and said carry input.
Claim 2
Dependent Claim
406 words
A floating point adder circuit using neural network which adds two binary numbers of K(M+N) bit that consist of an M-bit exponent part and an N-bit mantissa part comprising; a comparator for comparing the exponent magnitude of said two binary numbers by comprising first input synapses for connecting a first power supply voltage to a line as the connecting strength of the weight of each bit according to a first exponent value of said M-bit; second input synapses for connecting a second power supply voltage to said line as the connecting strength of the weight of each bit according to a second exponent value of said M-bit; and a neuron which is stimulated or restrained according to the difference of the connecting strength between said first power supply voltage and said second power supply voltage that are connected to said line; an adder which add the mantissa parts of said two binary numbers by comprising input synapses for respectively connecting said second power supply voltage to a first line of N+1 numbers as the connecting strength of the weight of each bit according to a first mantissa value of said N-bit, a second mantissa value of said N-bit and a carry input of one-bit; first bias synapses for connecting said first power supply voltage to each first line as the connecting strength of a multiple of said input connecting strength sum; feedback synapses for connecting said second power supply voltage to a second line of N+1 numbers of each lower output bit as the connecting strength of the weight of upper output bit; second bias synapses for connecting said first power supply voltage to the least significant bit line among said second lines as the connecting strength which subtracts 0.5 from the connecting strength of said first bias synapse and to the other lines as the connecting strength which subtracts the weight of the bit from the connecting strength of each lower bit line, in order to bias said each second line connected to said feedback synapses; neurons which are stimulated or restrained by comparing the difference of the connecting strength of said first power supply voltage and said second power supply voltage which are respectively connected to said first line and said second line; and inverters for inverting the output of each of said neurons; and a subtractor comprising said adder and inverters for inverting some mantissa value of said N-bit and said carry input.
Claim 3
Dependent Claim
32 words
The floating point adder circuit using neural network as claimed in claim 1, wherein said neuron of said comparator is composed of two CMOS inverters which are connected in series each other.
Claim 4
Dependent Claim
35 words
The floating point adder circuit using neural network as claimed in claim 2, wherein said first input synapses and said second input synapses of said comparator are made of PMOS transistors and NMOS transistors, respectively.
Claim 5
Dependent Claim
50 words
The floating point adder circuit using neural network as claimed in claim 3, wherein said input synapses and said feedback synapses of said adder are composed of NMOS transistors, and each of said first and said second bias synapses is composed of the combination of PMOS transistor and NMOS transistor.
Claim 6
Dependent Claim
24 words
The floating point adder circuit using neural network as claimed in claim 4, wherein said neurons of said adder are composed of differential amplifiers.
Claim 7
Dependent Claim
31 words
The floating point adder circuit using neural network as claimed in claim 5, wherein the connecting strength of each synapse is made of the conductance value of PMOS and NMOS transistor.
Claim 8
Dependent Claim
34 words
The floating point adder circuit using neural network as claimed in claim 6, wherein said conductance value of said PMOS or NMOS transistor is determined by the geometrical aspect ratio (W/L) of MOS transistor.
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) | 1995-02-03 | 1995-08-05 | 1996-02-04 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1999-02-03 | 1999-08-05 | 2000-02-04 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2003-02-03 | 2003-08-05 | 2004-02-04 | $7700.00 | Window Closed / Lapsed |