Awakening, looking ahead in 2026

Hapy New Year! I am planning to be more active here in 2026. I am still quite busy at SemiAnalysis and with occasional other consulting, but the ideas for Poratbo are still stacked up and nagging me to write.

There are a few of you who were interested in cryogenic computing. I recently noticed that DARPA is looking for proposals in this area. DARPA is an excellent way to fund a project, the main drawback is you need to be something they consider to be interesting. If your project interests them and they decide to give seed money for you to pursue your idea, there are few strings attached, minimal bureaucracy. So, it may be worth your time to interact with them if you want to explore cryogenic (but conventional) computing.

DARPA’s request for proposals begins here: https://www.darpa.mil/research/programs/low-temperature-logic-technology

You should read up on how to make a proposal to DARPA. They have their special style which you need to follow. You need to be concise, a proposal is only a few pages. None of those pages should be waffle. This is definitely a case where Mark Twain’s observation “I would have written a shorter letter, but I did not have the time” should caution you not to mistake brevity for easy.

The very cool low energy frontier

·

March 25, 2024

The Pipes has been an absorbing project. The design seems competitive design and estimates line up with real chips. It was a surprise to find that the silicon seems heat limited, and I’m not even using rules that map to the densest products now possible.

I have been using agents a lot for my work, and this is one of the things that convinces me to get back into writing this blog. Agents have been advancing rapidly. I get things done in an hour now which just caused frustration with an agent mere months ago. I need to run a competition every few weeks to see which agent is at the head of the pack, because it changes significantly.

I expect this will allow me to turn some of my idea queue into useful articles more efficiently. I expect to look at tensor units at the low level, investigating how they operate with low energy per operation. I also plan to explore the math of microscaled floating point (whether OCP MX, or Nvidia’s slightly diffent version) and see how it can be made stable and usefully accurate, matching it to efficient tensor units.

The gravity telescope has been on hold for a while because it proved very tricky to solve the orbit dynamics precisely. I will be working on that again. Last time it got no help from AI, but agents have improved. I will probably write progress reports on it rather than waiting until it is all finished.

I also want to go back and revisit the SRAM cell. I realized there is a way to make it 1/8th smaller, but more important I want to explore some of the stability and leakage issues that are key to successful SRAM. It looks like I may have a way to access some pro-quality simulation tools for this.

For the few who have kept chipping in with subscription, thank you very much!

I did open new blog posts to everyone beginning in 2025 and plan to keep it that way. There are a couple of thousand readers now. Do not be shy to comment or send messages, I can improve from your feedback and am especially happy to learn new things from readers.

Happy new year!