Siemens puts overnight design sweeps to work on 3D chips
Illustrative image: Unsplash contributor on Unsplash
Siemens’ new agent-driven workflow can spend eight to ten hours exploring a 3D IC design, coordinating electrical and thermal analysis across hundreds of design points. The scale of that exploration is the most interesting part of the announcement. It offers a way to examine combinations of power, materials and cooling assumptions that a manually managed programme might never reach.
The Fuse EDA AI Agent, integrated with Intelligence Center X, coordinates nine stages across seven tools. Siemens describes 50 to 100 reasoning steps during a cycle covering between 200 and more than 500 design points.
Its reported gains include up to 50 times greater exploration coverage, up to 95% less engineering effort and up to 99% less setup effort. These measure different things. A reduction in setup labour does not imply the same reduction in solver runtime or the time needed to complete a chip.
The engineering problem is coupled. Temperature affects electrical behaviour, electrical activity determines heat generation and a change in one part of a stacked assembly can alter conditions elsewhere. Analysing each discipline once, in isolation, can leave a design apparently acceptable until those interactions are considered.
Calibre 3DThermal already provides part of the foundation. It uses a customised Simcenter Flotherm solver and produces thermal information for electrical simulation. Models can develop from an early feasibility study into a detailed assembly incorporating metallisation and other physical features as the design matures.
Early exploration is therefore valuable even before every physical detail is available. It can expose a sensitive floorplan or a material choice that deserves closer examination. More detailed analysis can then concentrate on the alternatives that remain plausible.
Cooling assumptions need equally close attention. Siemens supports model exchange between Calibre 3DThermal and Flotherm so that die information can feed system analysis and realistic system boundary conditions can return to the chip model.
That exchange gives the overnight sweep a useful purpose: finding out whether an attractive package arrangement still works when the cooling conditions change. Running more cases makes that sensitivity easier to see. It also makes a poor assumption capable of influencing far more results.
