IP Library Granted Patent US 12,386,621
Granted Patent B2
US 12,386,621 · App. 18/517,031 · Granted Aug 12, 2025

AI synaptic coprocessor

Inventors: David Sherwood (Marriottsville, MD); Terry A. Higbee (Conifer, CO)
Assignee: COGNITIVE SCIENCE & SOLUTIONS, INC.
G06F9/30152G06F7/57G06F13/28G06N3/063
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,386,621
App. No.
18/517,031
Granted
Aug 12, 2025
Kind
B2
Abstract

A coprocessor may include a memory configured to store a plurality of Very Long Data Words, each as a test Very Long Data Word (VLDW) having a length in the range of about one thousand bits to one million or more bits and containing encoded information that is distributed across the length of the VLDW. A processor generates search terms and a processing logic unit receives a test VLDW from the memory, receives a search term from the processor, and computes a Boolean inner product between the search term and the test VLDW read from memory indicative of the measure of similarity between the test VLDW and the search term. Optionally, buffers within logic circuits of processing pipelines may receive the test VLDWs.

Claims (29)

1. A coprocessor, comprising:

a memory configured to store a plurality of Very Long Data Words, each comprising a test Very Long Data Word (VLDW) having a length in the range of about one thousand bits to one million or more bits and containing encoded information that is distributed across the length of the VLDW;

a processor configured to generate search terms; and

a processing logic unit configured to:

receive a plurality of test VLDWs from the memory;

receive a search term from the processor; and

compute Boolean inner products between the search term and the plurality of test VLDWs read from memory indicative of the measure of similarity between the plurality of test VLDWs and the search term.

2. The coprocessor of claim 1 , wherein the processing logic unit comprises one or more pipeline Boolean logic circuits that compute the Boolean inner product.

3. The coprocessor of claim 1 , further comprising a buffer configured to store the Boolean inner products resulting from the computation between each search term and the test VLDWs together with an address in memory from which the test VLDWs were read.

4. The coprocessor of claim 1 , further comprising a Direct Memory Access (DMA) controller connected to the processor and memory and configured to address and control the transfer of test VLDWs from memory to the processing logic unit.

5. The coprocessor of claim 1 , wherein said processor includes a conventional CPU interface for communicating with external devices, wherein said processor is configured receive VLDW words as a plurality of 64-bit words via the conventional CPU interface and reformat the 64-bit words into a test VLDW having a length of about one thousand bits to at least one million bits.

6. The coprocessor of claim 1 , comprising an external sensor connected to the coprocessor and configured to generate sensor data, said processor is configured to receive the sensor data and generate a sensed data VLSW from the sensor data, wherein said processing logic unit is configured to compare the sensed data VLDW to a search term.

7. The coprocessor of claim 1 , wherein said processor is configured to generate a plurality of test VLDWs, and said processing logic unit includes a processing buffer into which the plurality of test VLDWs are buffered.

8. The coprocessor of claim 1 , wherein said processor includes a serial interface and digital logic, wherein said serial interface passes serial data to said digital logic to reformat the serial data into very long data words.

9. A coprocessor, comprising:

a memory configured to store a plurality of Very Long Data Words, each comprising a test Very Long Data Word (VLDW) having a length in the range of about one thousand bits to one million or more bits and containing encoded information that is distributed across the length of the VLDW;

a processor configured to generate search terms; and

a processing logic unit configured to:

receive a plurality of test VLDWs from the memory;

receive a search term from the processor; and

compute successive Boolean inner products between the search term and the plurality of test VLDWs read from memory indicative of the measure of similarity between the plurality of test VLDWs and the search term.

10. The coprocessor of claim 9 , wherein said processing logic unit is configured to accumulate the successive Boolean inner products.

11. The coprocessor of claim 9 , further comprising a buffer configured to store the Boolean inner products resulting from the computation between each search term and the test VLDWs together with an address in memory from which the test VLDWs were read.

12. The coprocessor of claim 9 , further comprising a Direct Memory Access (DMA) controller connected to the processor and memory and configured to address and control the transfer of test VLDWs from memory to the processing logic unit.

13. The coprocessor of claim 9 , wherein said processor includes a conventional CPU interface for communicating with external devices, wherein said processor is configured receive VLDW words as a plurality of 64-bit words via the conventional CPU interface and reformat the 64-bit words into a test VLDW having a length of about one thousand bits to at least one million bits.

14. The coprocessor of claim 9 , comprising an external sensor connected to the coprocessor and configured to generate sensor data, said processor is configured to receive the sensor data and generate a sensed data VLSW from the sensor data, wherein said processing logic unit is configured to compare the sensed data VLDW to a search term.

15. The coprocessor of claim 9 , wherein said processor is configured to generate a plurality of test VLDWs, and said processing logic unit includes a processing buffer into which the plurality of test VLDWs are buffered.

16. The coprocessor of claim 9 , wherein said processor includes a serial interface and digital logic, wherein said serial interface passes serial data to said digital logic to reformat the serial data into very long data words.

17. The coprocessor of claim 9 , wherein the processor is configured to perform calculations at a clock rate and compute Boolean inner products between the search term and test VLDWs at a latency to obtain results after multiple clocks.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 5, 2023
From: SHERWOOD, DAVID; HIGBEE, TERRY A.
To: COGNITIVE SCIENCE & SOLUTIONS, INC.
Reel/Frame 065759/0268 →
Continuity (4)
Continuation In Part 18168597 · Feb 14, 2023
Continuation 17242374 · Apr 28, 2021
Provisional Application 63124923 · Dec 14, 2020
Related Publication 20240095032A1 · Mar 21, 2024
References Cited (47)
US 3962685A · Belle Isle · 1976 [cited by examiner]
US 5359697A · Smith et al. · 1994 [cited by applicant]
US 6014672A · Satoh · 2000 [cited by examiner]
US 6584223B1 · Shiiyama · 2003 [cited by examiner]
US 7129954B2 · Sekine · 2006 [cited by examiner]
US 7130849B2 · Yayoi · 2006 [cited by examiner]
US 7739257B2 · Ellis · 2010 [cited by examiner]
US 8429153B2 · Birdwell · 2013 [cited by examiner]
US 8601244B2 · Egger et al. · 2013 [cited by applicant]
US 9659248B1 · Barbosa · 2017 [cited by examiner]
US 9971965B2 · Amir · 2018 [cited by examiner]
US 10255323B1 · Guo · 2019 [cited by examiner]
US 10303717B2 · Ishii · 2019 [cited by examiner]
US 10489468B2 · Lytkin · 2019 [cited by examiner]
US 10860641B2 · Liu · 2020 [cited by examiner]
US 10963738B2 · Geraci · 2021 [cited by examiner]
US 11023477B2 · Fan · 2021 [cited by examiner]
US 11062180B2 · Guo · 2021 [cited by examiner]
US 11080607B1 · Demtchenko · 2021 [cited by examiner]
US 11113291B2 · Gotmanov · 2021 [cited by examiner]
US 11218695B2 · Park · 2022 [cited by examiner]
US 11328733B2 · Mosayyebpour Kaskari · 2022 [cited by examiner]
US 11392596B2 · Wu · 2022 [cited by examiner]
US 11409752B1 · Qadrud-Din et al. · 2022 [cited by applicant]
US 11544309B2 · Toyoshiba · 2023 [cited by examiner]
US 11599360B2 · Sherwood · 2023 [cited by examiner]
US 11704312B2 · Pasternack · 2023 [cited by examiner]
US 11704714B2 · Hiranandani · 2023 [cited by examiner]
US 11868776B2 · Sherwood · 2024 [cited by examiner]
US 20120321125A1 · Choi · 2012 [cited by examiner]
US 20150324688A1 · Wierzynski · 2015 [cited by examiner]
US 20190065913A1 · Kidera · 2019 [cited by examiner]
US 20190122119A1 · Husain · 2019 [cited by applicant]
US 20190205741A1 · Gupta et al. · 2019 [cited by applicant]
US 20200167654A1 · Guo et al. · 2020 [cited by applicant]
US 20210365515A1 · Jiang · 2021 [cited by examiner]
US 20210374353A1 · Zhang · 2021 [cited by examiner]
US 20220188116A1 · Sherwood · 2022 [cited by examiner]
US 20220414445A1 · Rasch · 2022 [cited by examiner]
US 20230205529A1 · Sherwood · 2023 [cited by examiner]
CN 106776782A · 2017 [cited by examiner]
CN 111344697A · 2020 [cited by examiner]
CN 111881334A · 2020 [cited by examiner]
CN 114398882A · 2022 [cited by examiner]
WO WO2019085236A1 · 2019 [cited by examiner]
Bytyn et al., “An Application-Specific VLIW Processor with Vector Instruction Set for CNN Acceleration,” 2019 IEEE International Symposium on Circuits and Systems (ISCAS); May 26-29, 2019; 6 pages. See Priority U.S. App… [cited by applicant]
Hu et al., “Dot-Product Engine for Neuromorphic Computing: Programming 1T1M Crossbar to Accelerate Matrix-Vector Multiplication,” Hewlett Packard Labs; HPE-2016-23; Hewlett Packard Enterprises; Mar. 3, 2016; pp. 1-7. Se… [cited by applicant]