Granted Patent B2

US 12,555,025 · App. 17/095,976 · Granted Feb 17, 2026

Inventors: Gandhi Karuna K T (Chennai, IN); Veerendra Prasad Nettem (Chennai, IN); Nataraj Pinakapani (Chennai, IN); Saravanan K (Chennai, IN)

Assignee: HCL Technologies Limited

G06N20/00G06F7/57G06F30/367

Abstract

Claims (62)

1 . A method for integrating Field Programmable Analog Array (FPAA) with Artificial Intelligence (AI), the method comprising:

automatically creating, by an AI model, a function by auto-connecting a first set of computation elements from a plurality of computational elements in the FPAA, in response to receiving an input, wherein the function generates an output associated with the input, and wherein the plurality of computational elements comprises at least one of a divider translinear circuit, an adder translinear circuit, a multiplier translinear circuit, a subtractor translinear circuit, or a logarithmic translinear circuit;

receiving, by the AI model, a feedback comprising a first accuracy level associated with the output; and

automatically adjusting, by the AI model, at least one of a plurality of control parameters to modify the function to generate an adjusted output corresponding to the input, based on the first accuracy level associated with the output,

wherein the plurality of control parameters corresponds to external environmental factors affecting performance of the FPAA in generating the output, and

wherein automatically adjusting the at least one of the plurality of control parameters comprises auto-connecting a second set of computation elements from the plurality of computational elements to compensate for the external environmental factors;

reconfiguring the FPAA, by the AI model, to apply a modified function based on the control parameters, wherein the control parameters are generated in response to ambient lighting conditions to adapt the configuration of the FPAA and maintain the output accuracy,

wherein the FPAA is a Field Programmable Translinear Array (FPTA); and

reprogramming the FPTA, by the AI model, to perform a task comprising identifying a visual attribute of an object, wherein the reprogramming maintains identification accuracy under non-ideal environmental conditions.

2 . The method of claim 1 , wherein the FPAA comprises each of the plurality of computational elements, and wherein each of the plurality of computational elements is configured to perform one of subtraction, addition, multiplication, division, or logarithmic computation.

3 . The method of claim 2 , wherein the FPAA and the AI model are implemented as at least one of a single integrated circuit, a co-processor, or an AI accelerator embedded in an integrated circuit.

4 . The method of claim 1 , further comprises training the AI model, wherein the training comprises:

receiving, by the AI model, a set of input data and an associated output data;

creating, by the AI model, a plurality of functions for each of the set of input data based on each of the set of input data and the output data; and

applying, by the AI model, each of the plurality of functions on the FPAA.

5 . The method of claim 4 , wherein creating the plurality of functions, further comprises:

receiving, by the AI model, a feedback based on an intermediate output generated corresponding to each of the set of input data;

comparing, by the AI model, the intermediate output, for each of the set of input data, with the output data to determine a second accuracy level of the intermediate output; and

incrementally adapting, by the AI model, the function based on the second accuracy level.

6 . A single integrated circuit, the single integrated circuit comprising:

a Field Programmable Analog Array (FPAA) comprising a plurality of computational elements; and

an Artificial Intelligence (AI) model including an AI learning engine, the AI model configured to:

automatically create a function by auto connecting a first set of computation elements from a plurality of computational elements in the FPAA, in response to receiving an input, wherein the function generates an output associated with the input, and wherein the plurality of computational elements comprises at least one of a divider translinear circuit, an adder translinear circuit, a multiplier translinear circuit, a subtractor translinear circuit, or a logarithmic translinear circuit;

receive a feedback comprising the first accuracy level associated with the output; and

automatically adjust at least one of a plurality of control parameters to modify the function to generate an adjusted output corresponding to the input, based on the first accuracy level associated with the output,

wherein the plurality of control parameters corresponds to external environmental factors affecting performance of the FPAA in generating the output, and

wherein automatically adjusting the at least one of the plurality of control parameters comprises auto-connecting a second set of computation elements from the plurality of computational elements to compensate for the external environmental factors;

reconfigure the FPAA to apply a modified function based on the control parameters, wherein the control parameters are generated in response to ambient lighting conditions to adapt the configuration of the FPAA and maintain the output accuracy,

wherein the FPAA is a Field Programmable Translinear Array (FPTA); and

reprogram the FPTA to perform a task comprising identifying a visual attribute of an object, wherein the reprogramming maintains identification accuracy under non-ideal environmental conditions.

7 . The single integrated circuit of claim 6 , wherein the FPAA comprises each of the plurality of computational elements, and wherein each of the plurality of computational elements is configured to perform one of subtraction, addition, multiplication, division, or logarithmic computation.

8 . The single integrated circuit of claim 7 , wherein the FPAA and the AI model are implemented as at least one of a single Integrated Circuit, a co-processor, or an AI accelerator embedded in an integrated circuit.

9 . The single integrated circuit of claim 6 , wherein the AI learning engine is configured to train the AI model based on a log database, and wherein training comprises:

receiving, by the AI model, a set of input data and an associated output data;

creating, by the AI model, a plurality of functions for each of the set of input data based on each of the set of input data and the output data; and

applying, by the AI model, each of the plurality of functions on the FPAA.

10 . The single integrated circuit of claim 9 , wherein the AI learning engine creates the plurality of functions by:

receiving, by the AI model, a feedback based on an intermediate output generated corresponding to each of the set of input data;

comparing, by the AI model, the intermediate output, for each of the set of input data, with the output data to determine the accuracy level of the intermediate output; and

incrementally adapting, by the AI model, the function based on the determined accuracy level.

11 . An Artificial Intelligence (AI) accelerator, the system comprising:

a Field Programmable Analog Array (FPAA) comprising a plurality of computational elements; and

an AI model including an AI learning engine, the AI model configured to:

automatically create a function by auto connecting a first set of computation elements from a plurality of computational elements in the FPAA, in response to receiving an input, wherein the function generates an output associated with the input, and wherein the plurality of computational elements comprises at least one of a divider translinear circuit, an adder translinear circuit, a multiplier translinear circuit, a subtractor translinear circuit, or a logarithmic translinear circuit;

receive a feedback comprising the first accuracy level associated with the output; and

automatically adjust at least one of a plurality of control parameters to modify the function to generate an adjusted output corresponding to the input, based on the first accuracy level associated with the output,

wherein the plurality of control parameters corresponds to external environmental factors affecting performance of the FPAA in generating the output, and

wherein automatically adjusting the at least one of the plurality of control parameters comprises auto-connecting a second set of computation elements from the plurality of computational elements to compensate for the external environmental factors:

reconfigure the FPAA to apply a modified function based on the control parameters, wherein the control parameters are generated in response to ambient lighting conditions to adapt the configuration of the FPAA and maintain the output accuracy,

wherein the FPAA is a Field Programmable Translinear Array (FPTA); and

reprogram the FPTA to perform a task comprising identifying a visual attribute of an object, wherein the reprogramming maintains identification accuracy under non-ideal environmental conditions.

12 . The AI accelerator of claim 11 , wherein the FPAA comprises each of the plurality of computational elements, and wherein each of the plurality of computational elements is configured to perform one of subtraction, addition, multiplication, division, or logarithmic computation.

13 . The AI accelerator of claim 12 , wherein the FPAA and the AI model are implemented as at least one of a single Integrated Circuit, a co-processor, or an AI accelerator embedded in a large integrated circuit.

14 . The AI accelerator of claim 11 , wherein the AI learning engine is configured to train the AI model based on a log database, and wherein training comprises:

receiving, via the AI model, a set of input data and an associated output data;

creating, via the AI model, a plurality of functions for each of the set of input data based on each of the set of input data and the output data; and

applying, via the AI model, each of the plurality of functions on the FPAA.

15 . The AI accelerator system of claim 11 , wherein the AI learning engine creates the plurality of functions by:

receiving, via the AI model, a feedback based on an intermediate output generated corresponding to each of the set of input data;

comparing, via the AI model, the intermediate output, for each of the set of input data, with the output data to determine the accuracy level of the intermediate output; and

incrementally adapting, via the AI model, the function based on the determined accuracy level.

16 . The AI accelerator of claim 11 , wherein the AI accelerator is embedded in an integrated circuit.

Assignments (1)

Continuity (1)

References Cited (20)