AI learns the “dark art” of RFIC design

What did Princeton researchers actually do with AI chip design?

A team at Princeton, led by engineering professor Kaushik Sengupta, used reinforcement learning and diffusion models (the same family of AI behind image generators) to design radio frequency integrated circuits, or RFICs. This is the branch of chip design engineers have long called a "dark art" because it relies on decades of hand tuned intuition rather than fixed rules.

The AI produced power amplifier layouts that look nothing like a human would draw them, some resembling QR codes rather than the neat, symmetrical patterns engineers favour. Despite the odd appearance, they beat the best existing silicon based circuits on bandwidth, output power and efficiency across the demanding 30 to 100 GHz frequency range used in 5G and radar systems.

Training the system took a few days to a week. After that, it could generate new working circuit designs in minutes, work that previously took specialist engineers days or weeks. A separate AI emulator built with convolutional neural networks now predicts circuit behaviour in milliseconds instead of the minutes or hours traditional simulators need, and a full diffusion model design cycle runs in about 6 minutes.

Why should a UK small business owner care about chip design AI?

Because it is a clean signal of how far AI has moved past writing emails and captions. If reinforcement learning can now outperform veteran engineers at a physics constrained, decades old specialist craft, the far simpler, more repetitive tasks inside a typical small business are well within reach already, not in some distant future.

This matters because UK small business AI adoption has moved fast. According to the British Chambers of Commerce and Atos, 54% of UK SMEs used AI in some form in 2026, up from 35% in 2025, 25% in 2024 and 23% in 2023. Your competitors are more likely than not already using it.

If I haven't adopted AI yet, am I already behind?

Not necessarily. Adoption has jumped, but results have not kept pace: only 12% of AI using UK businesses report a measurable revenue increase so far, even though marketing is the top use case, named by 72% of adopters. Using AI and benefiting from it are two different things.

The gap usually comes down to how AI is used. Businesses bolting a chatbot onto a broken booking process, or using AI to write social posts nobody reads because there is no real website behind them, are adding activity without adding value. The Princeton result is useful precisely because it worked: they aimed the AI at one narrow, well defined problem with a clear success measure (bandwidth, power, efficiency) rather than a vague hope that "AI will help".

What should a small business actually do with this news?

Pick one specific, measurable job for AI to do, on infrastructure you own, and check the result. Do not adopt AI as a general gesture.

• Use AI to draft Google Business Profile posts and review replies, but keep the profile itself verified and fully optimised first. One Leeds salon went from no online presence to over 40 monthly calls from Google Maps within 3 weeks of properly claiming and completing its profile, before any AI was involved.
• Let AI handle first line customer replies or booking confirmations on your own website or system, not inside a rented platform you could lose access to.
• Measure one number before and after (calls, bookings, orders), the same way the Princeton team measured bandwidth and efficiency, so you know if it actually worked rather than assuming it did.

What is Braynex Services' take on this?

The lesson from a Princeton lab is the same one we see repeatedly with local businesses: AI is only as good as the infrastructure it is plugged into. A brilliant AI generated design still needs a real chip to sit on. A brilliant AI generated post or reply still needs a real website, booking system and data you actually own to sit behind it.

We have watched a nail salon paying Fresha roughly 1,800 a month in commission move to its own booking system for 35 a month, saving around 21,000 a year, and a Manchester restaurant lift online orders by 34% within 4 weeks simply by replacing a PDF menu with a live web one. None of that needed cutting edge AI. It needed ownership.

Our other prediction is that being recommended by AI assistants such as ChatGPT, Gemini and Google AI Overviews is becoming the new local SEO. Those tools pull from structured, accurate, owned information, not from a Facebook page or a Linktree that could disappear tomorrow. Get the foundations right first, then layer AI on top, not the other way round.

If you want a clear picture of where your business is leaking calls, bookings or visibility, and whether AI would genuinely help or just add noise, book a free audit at braynexservices.com.