LLM‑Powered Agent Automates Analog/Mixed‑Signal I/O Design, Cutting Turnaround to Minutes

Global: LLM‑Powered Agent Automates Analog/Mixed‑Signal I/O Design, Cutting Turnaround to Minutes

Researchers have introduced AMS‑IO‑Agent, a domain‑specialized large language model (LLM) agent that converts natural‑language design intent into industrial‑grade analog and mixed‑signal (AMS) integrated‑circuit I/O subsystems. The work, posted to arXiv in December 2025, aims to streamline the traditionally labor‑intensive I/O ring design process and to demonstrate a practical human‑agent collaboration in silicon production.

Framework Overview

According to the preprint, the framework links user‑provided intent with two core capabilities: a structured knowledge base that encodes reusable constraints and design conventions, and a design‑intent structuring module that translates ambiguous requests into verifiable logic steps using JSON and Python as intermediate representations.

Knowledge Base and Intent Structuring

The knowledge base captures industry‑standard rules for AMS I/O, enabling the agent to reuse constraint sets across projects. Meanwhile, the intent‑structuring component parses free‑form specifications, generates a step‑by‑step plan, and produces code snippets that can be directly fed into electronic‑design‑automation (EDA) tools.

Benchmark Performance

The authors present AMS‑IO‑Bench, a benchmark that evaluates wire‑bond‑packaged AMS I/O ring automation. On this benchmark, AMS‑IO‑Agent achieves a design‑rule‑check (DRC) and layout‑versus‑schematic (LVS) pass rate of over 70%, and reduces design turnaround time from several hours to a matter of minutes, outperforming a baseline LLM without the specialized framework.

Silicon Validation

To validate the approach, an agent‑generated I/O ring was fabricated in a 28 nm CMOS tape‑out. Post‑fabrication testing confirmed that the ring met the specified electrical performance, providing concrete evidence that the agent’s outputs can be used directly in silicon without manual re‑engineering.

Implications for AMS Design

The authors claim this is the first reported instance of a human‑agent collaborative workflow in which an LLM‑based agent completes a non‑trivial AMS IC subtask with deliverables that proceed straight to tape‑out. If widely adopted, such agents could accelerate design cycles, lower engineering costs, and open new avenues for integrating AI into hardware development pipelines.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.