The hardware conversation online lives in two places. Vendor decks that overstate the numbers, and forum threads that undercount them. Neither is a research corpus. GIKSN Research is our attempt to give hardware a place where the numbers are cited, the trade-offs are named and the assumptions are documented before the argument gets made.

What hardware work here means

A hardware paper on GIKSN treats specifications as claims and claims as arguments. TOPS is not enough. We want to know at what precision, at what batch size, at what sequence length and with what memory bandwidth utilisation. If a number is estimated we say so. If a design axis loses on a workload it was not built for we say that too. The point is that the paper survives its own numbers a year later.

What we cover

Silicon architecture, accelerators for transformer inference and training, embedded systems for autonomous applications, sensors and the packaging around them. We also cover the tooling on top of the hardware because the tooling decides whether the hardware ships or gathers dust. A verilog cell library without a synthesis flow argument is not a hardware paper. It is a set of files.

Why publish in the open

Hardware research suffers when it is only written for one audience. Vendors write papers for buyers. Academics write papers for tenure committees. GIKSN aims to write for the small group of people who will actually build on the paper. That means the failure mode section is the most important section, and the comparison baseline is stated in terms the reader can reproduce.

The cross-sector view

An AI paper without hardware honesty tends to overstate what the model can do at scale. A distributed systems paper without hardware honesty tends to assume interconnect that does not exist. A deeptech paper without hardware honesty tends to ignore the fabrication pipeline. The archive is designed so that a hardware paper can be linked directly to the AI or DS or DT paper that leans on it, and the discussion happens across the boundary rather than inside a silo.

What we are shipping

We do not tape out chips in a garage. What we ship from the hardware bench are open reference designs, characterisation tools and honest evaluations of publicly available parts. When the lab has a product ready for the products page it goes up with reproducible numbers, license terms and a link to the paper that argues for it.

Reading and contributing

Reading is open. Commentary is welcomed. Contribution to active hardware projects is gated because the work often needs sustained expertise, and the About-page application is the path in. Accepted contributors join the private working channels on Telegram.