Chip arithmetic is built from concrete multiplier, adder, and approximation structures, not from the word "AI" by itself.

The post points readers toward the arithmetic layer: Wallace trees, Han-Carlson adders, fused multiply-add units, Newton-Raphson approximation, and the transistor-level blocks behind GPU, FPGA, and ASIC datapaths.

LinkedIn postPosted 2023-11-09.

Included: the original post and a compact framing around chip arithmetic primitives.

Original Post

If you don't know how to multiply and add numbers, you don't know anything. Learn, how IBM, Nvidia, Intel, AMD multiply numbers, transistor-by-transistor. All patents are open and old. Make your ASIC. Innovate.

If you don't know what Wallace tree and Han-Carlson are, you don't know anything. Make your own FMA. The school arithmetic they taught you is too slow. Learn how Newton-Raphson can approximate any division and exponentiation to floating point precision fast without actual cycles and counters. Put your entire neural network on chip. Don't just train a LLaMA. Make your own.

#chips #math #ai #innovation #gpu #gpgpucompute #fpga #asic #nvidia #tsmc #verilog #gdsii #diy #gpt