Agentic AI Solutions for Efficient 6G Hardware Design

Paper: Toward Agentic AI for the 6G Physical Layer: From Algorithm Discovery to Verified Baseband Silicon

Authors: Eric Mittag, Lukas Brechtel, Maximilian Stark, Hans D. Schotten

Executive Summary

This research explores the intersection of artificial intelligence and hardware design, specifically in the context of developing algorithms for 6G technology. As the demand for faster and more efficient communication systems grows, traditional methods of designing hardware are becoming increasingly inadequate. The study presents a new workflow that utilizes a machine learning model, referred to as a coding agent, to generate and optimize algorithms for commercial baseband systems-on-chip (SoCs).

The core of the research focuses on the limitations of existing silicon technology, which imposes strict memory and numeric constraints that differ from those typically encountered in graphics processing units (GPUs). The proposed workflow allows the coding agent to implement algorithms directly onto these embedded systems, optimizing them down to the level of vector assembly. This process ensures that the algorithms are not only theoretically sound but also practical for real-world hardware.

The researchers demonstrate this workflow using an orthogonal time frequency space (OTFS) detector portfolio on the open 6G4All platform. They successfully develop twelve kernels that operate precisely on the silicon, achieving optimized performance ranging from 18,000 to 405,000 cycles per frame, translating to processing times between 30 and 660 microseconds. Notably, the workflow led to the creation of a new detector that operates with 13.9% fewer cycles compared to a previously discovered detector while maintaining the same error rate. This improvement highlights the ability of the coding agent to identify cost-effective solutions that traditional methods might overlook.

The significance of this research lies in its potential applications beyond just the OTFS detector. The workflow can be adapted to explore various implementation spaces under real hardware constraints, potentially benefiting a range of industries that rely on advanced communication technologies. By measuring the cost of every design choice rather than making assumptions, the approach could lead to more efficient and economically viable hardware solutions.

The study relies on simulated benchmarks rather than real-world applications at this stage, suggesting that while the findings are promising, further validation in practical settings will be crucial. This research could pave the way for faster development cycles in hardware design, particularly as we approach the era of 6G and beyond. By integrating AI into the design process, companies may find new efficiencies and capabilities that were previously unattainable, fostering innovation in telecommunications and related fields.

Academic Abstract

6G will be designed faster than it can be built: AI-driven search already proposes physical-layer algorithms at machine speed, while carrying a single one onto embedded baseband silicon remains specialist work. That silicon imposes hard memory and numeric limits, but also exposes cost structures a GPU lacks, and a design flow that never runs on the target sees neither. We present an agentic workflow in which an LLM coding agent implements algorithms on a commercial baseband SoC, optimizes them down to agent-written vector assembly, and validates them on silicon. The agent is guided by a knowledge base of measured platform facts, checked at every step against evaluators it is instructed never to edit, and metered per phase. We showcase the workflow on an AI-discovered orthogonal time frequency space (OTFS) detector portfolio on the open 6G4All platform. It yields twelve kernels that run cycle-exactly on silicon, their optimized versions at 18k-405k cycles (30-660 $\mu$s) per frame. It also yields a new detector shaped by the chip's measured costs that needs 13.9% fewer cycles than the port of the discovered detector while matching its error rate, with a measured edge our evaluation cannot statistically resolve, a structure that a hardware-blind search objective had no reason to find. What is not specific to OTFS is the workflow: an LLM agent exploring an implementation space under real hardware limits, with the cost of every choice measured rather than assumed.