# US 2026/0310867 A1 · Circuit topology synthesis using large language models and reinforcement learning

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0310867 A1
- Document ID: US20260310867A1
- Status: Published application
- Kind: A1
- Application number: 19/097,913
- Filed: 2025-04-02
- Published: 2026-10-08
- Earliest claimed: 2025-04-02
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Prashanth Vijayaraghavan; Luyao Shi; Xin Zhang
- Web page: https://patentgrep.com/patent/US20260310867A1

## CPC

- G06F 30/327: Logic synthesis; Behaviour synthesis, e.g. mapping logic, HDL to netlist, high-level language to RTL or netlist
- G06N 3/092: Reinforcement learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Examples described herein provide for synthesis of an analog circuit using a large language model (LLM) and reinforcement learning (RL). Aspects include receiving an instruction prompt specifying a plurality of design constraints, including a component pool for the analog circuit, a minimum efficiency for the analog circuit, and output voltage for the analog circuit, inputting the instruction prompt into a fine-tuned LLM, and receiving, from the fine-tuned LLM, a circuit topology. Aspects also include applying RL to the circuit topology, wherein the RL includes defining a reward function that evaluates the circuit topology based on criteria and calculating a reward score for the circuit topology, iteratively refining, by the fine-tuned LLM, the circuit topology by inputting the circuit topology, the reward score, and the instruction prompt into the fine-tuned LLM, and based on a determination that the circuit topology meets the plurality of design constraints, outputting the circuit topology.

## Claims

### Claim 1

A computer-implemented method for synthesis of an analog circuit using a large language model (LLM) and reinforcement learning (RL), comprising:
receiving an instruction prompt specifying a plurality of design constraints, including a component pool for the analog circuit, a minimum efficiency for the analog circuit, and output voltage for the analog circuit;
inputting the instruction prompt into a fine-tuned LLM;
receiving, from the fine-tuned LLM, a circuit topology;
applying reinforcement learning to the circuit topology, wherein the reinforcement learning includes defining a reward function that evaluates the circuit topology based on one or more criteria and calculating a reward score for the circuit topology;
iteratively refining, by the fine-tuned LLM, the circuit topology by inputting the circuit topology, the reward score, and the instruction prompt into the fine-tuned LLM; and
based on a determination that the circuit topology meets the plurality of design constraints, outputting the circuit topology.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein the fine-tuned LLM is created by fine tuning a pre-trained LLM using supervised learning on a dataset comprising pairs of instruction prompts and corresponding circuit topology netlists.

### Claim 3 (depends on 2)

The computer-implemented method of claim 2, wherein the supervised learning includes performing incident encoding to transforming each entry in the circuit topology netlists into a language-like structure that includes an identification of each component and its connections.

### Claim 4 (depends on 1)

The computer-implemented method of claim 1, wherein applying reinforcement learning further includes tuning the fine-tuned LLM to maximize the reward function using policy gradient in reinforcement learning.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, wherein the circuit topology includes component configurations, connections, and parameters that define a functionality of a circuit.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, wherein reward function is a weighted function that is based on a circuit validity estimate, an efficiency estimate, and an expected output voltage estimate.

### Claim 7 (depends on 6)

The computer-implemented method of claim 6, wherein the circuit validity estimate is created based on a trained binary classifier that determines a validity of the circuit topology, the efficiency estimate is created by a regression model that estimates an efficiency of the circuit topology, and the expected output voltage estimate is created by a regression model that estimates an output voltage of the circuit topology.

### Claim 8

A system comprising:
a memory comprising computer readable instructions; and
a processing device for executing the computer readable instructions, the computer readable instructions controlling the processing device to perform operations comprising:
receiving an instruction prompt specifying a plurality of design constraints, including a component pool for an analog circuit, a minimum efficiency for the analog circuit, and output voltage for the analog circuit;
inputting the instruction prompt into a fine-tuned LLM;
receiving, from the fine-tuned LLM, a circuit topology;
applying reinforcement learning to the circuit topology, wherein the reinforcement learning includes defining a reward function that evaluates the circuit topology based on one or more criteria and calculating a reward score for the circuit topology;
iteratively refining, by the fine-tuned LLM, the circuit topology by inputting the circuit topology, the reward score, and the instruction prompt into the fine-tuned LLM; and
based on a determination that the circuit topology meets the plurality of design constraints, outputting the circuit topology.

### Claim 9 (depends on 8)

The system of claim 8, wherein the fine-tuned LLM is created by fine tuning a pre-trained LLM using supervised learning on a dataset comprising pairs of instruction prompts and corresponding circuit topology netlists.

### Claim 10 (depends on 9)

The system of claim 9, wherein the supervised learning includes performing incident encoding to transforming each entry in the circuit topology netlists into a language-like structure that includes an identification of each component and its connections.

### Claim 11 (depends on 8)

The system of claim 8, wherein applying reinforcement learning further includes tuning the fine-tuned LLM to maximize the reward function using policy gradient in reinforcement learning.

### Claim 12 (depends on 8)

The system of claim 8, wherein the circuit topology includes component configurations, connections, and parameters that define a functionality of a circuit.

### Claim 13 (depends on 8)

The system of claim 8, wherein reward function is a weighted function that is based on a circuit validity estimate, an efficiency estimate, and an expected output voltage estimate.

### Claim 14 (depends on 13)

The system of claim 13, wherein the circuit validity estimate is created based on a trained binary classifier that determines a validity of the circuit topology, the efficiency estimate is created by a regression model that estimates an efficiency of the circuit topology, and the expected output voltage estimate is created by a regression model that estimates an output voltage of the circuit topology.

### Claim 15

A computer program product for performing self-health checks in monitoring products, the computer program product comprising:
a set of one or more computer-readable storage media;
program instructions, collectively stored in the set of one or more storage media, for causing a processor set to perform the following computer operations:
receiving an instruction prompt specifying a plurality of design constraints, including a component pool for an analog circuit, a minimum efficiency for the analog circuit, and output voltage for the analog circuit;
inputting the instruction prompt into a fine-tuned LLM;
receiving, from the fine-tuned LLM, a circuit topology;
applying reinforcement learning to the circuit topology, wherein the reinforcement learning includes defining a reward function that evaluates the circuit topology based on one or more criteria and calculating a reward score for the circuit topology;
iteratively refining, by the fine-tuned LLM, the circuit topology by inputting the circuit topology, the reward score, and the instruction prompt into the fine-tuned LLM; and
based on a determination that the circuit topology meets the plurality of design constraints, outputting the circuit topology.

### Claim 16 (depends on 15)

The computer program product of claim 15, wherein the fine-tuned LLM is created by fine tuning a pre-trained LLM using supervised learning on a dataset comprising pairs of instruction prompts and corresponding circuit topology netlists.

### Claim 17 (depends on 16)

The computer program product of claim 16, wherein the supervised learning includes performing incident encoding to transforming each entry in the circuit topology netlists into a language-like structure that includes an identification of each component and its connections.

### Claim 18 (depends on 15)

The computer program product of claim 15, wherein applying reinforcement learning further includes tuning the fine-tuned LLM to maximize the reward function using policy gradient in reinforcement learning.

### Claim 19 (depends on 15)

The computer program product of claim 15, wherein the circuit topology includes component configurations, connections, and parameters that define a functionality of a circuit.

### Claim 20 (depends on 15)

The computer program product of claim 15, wherein reward function is a weighted function that is based on a circuit validity estimate, an efficiency estimate, and an expected output voltage estimate.
