In December 2021 Chipchain's engineers wrote the company's full-custom design flow into a getting-started course for engineers joining its design programs.

The course follows a chip from the first team meeting to packaged parts. Logic design, custom cells and software start in parallel, and the logic is proved on an FPGA before any layout. Then come a hand-crafted netlist, custom clock trees and manual place and route, the board design, full-flow simulation and verification, tape-out, and finally packaging and test.

Its central lesson is where full-custom design pays. Custom cells, netlists mapped by hand, an optimised clock tree, manual placement and a strengthened power grid, with the front end and the back end tuned together, cut the switching that does no useful work, and that raises the energy efficiency of every operation.

The order matters as much as the techniques. Proving the logic on an FPGA first means the hand-built netlist starts from a design that is known to be right, so the weeks spent on cells, clocks and placement go into speed and power rather than into fixing function.

Each technique in the course is also a service Chipchain offers on its own, from standard-cell library characterisation to custom clock trees, for a customer's datapath or compute array.