AI is limited as much by the hardware it runs on as by its models. Anandians Connections was founded as an AI and semiconductor company, and our research reaches down to the chip.
Where the research stands
AI accelerator silicon: two deep-tech semiconductor products in experimentation, with patents pending. ResearchPatent pendingOur plan is global production and public release once the patents are approved. Until then, this work is research, and we describe it as research.
The four areas
- Accelerator chips: a programmable logic device custom-built for specific AI applications. Research
- ASICs: application-specific integrated circuits designed for particular AI workloads. Research
- FPGAs: logic that can be reconfigured after manufacture, for fast iteration on AI designs. Research
- Chip architecture optimised for low latency, high throughput and power efficiency. Research
Why it matters
Each area trades flexibility for efficiency in a different way. An FPGA can be reconfigured after manufacture, which makes it the fast path for trying a design; an ASIC is fixed, and can be far more efficient at the one job it was designed for; an accelerator built on programmable logic sits between the two. Chip architecture decides how all of them balance latency, throughput and power.
We will publish more here as the work moves from research towards production — with its status, and nothing more than the evidence supports. See our technology.
Sources
- Anandians Connections — anandians.org home page, as archived (Internet Archive)
- Anandians Connections — semiconductor services page on anandians.com, as archived (Internet Archive)
Every figure in this article is recorded with its source and the date it was checked. More insights.