01
Semiconductors & compute
Where computation and state should live, and what silicon the workload actually needs.
New public research
China & U.S. AI Chip Atlas.
We are building a China-first, source-linked map of AI chips and the supply chains behind them, with a bounded U.S. comparison.
The current release tracks 55 product paths across 15 stages, from software and memory through fabrication, deep-ultraviolet and extreme-ultraviolet lithography, packaging, systems, qualification, and deployment. It keeps product facts, named supplier relationships, and unknowns separate.
The separate progress ledger preserves 38 dated milestones across 28 exact products at the readiness gate each source supports. Progress is not silently promoted into matched performance or qualified volume.
Research status: This is working research, not a finished or fully verified database. It is incomplete and may be wrong or stale. Check the cited source and review date before relying on any record.
Our goal
Research should end in something you can build, test, or decide.
Research should end in a better decision, experiment, product, or process.
Customer work and independent fab research can each raise a testable question. For a customer product, compare capable software, hardware, labs, and manufacturers first. For an owned fab process, define the workpiece, measurement, site, operator, cost, and stop decision before physical release.
We use existing specialists and facilities where they are strongest while building the knowledge, tooling, people, and process discipline required to own more of the manufacturing stack over time.
The goal is not a logo wall or a paper fab. The goal is repeatable capability.
What we study
What we research.
02
Materials & packaging
Heat, reliability, interconnect, substrates, bonding, and manufacturability.
03
Manufacturing & supply chain
Process, equipment, test, metrology, repair, sourcing, and repeat production.
04
Facilities & resources
Power, water, cooling, land, noise, workforce, and the constraints that decide whether infrastructure can actually be built.
Research collaboration
Bring a question we can test.
If you build chips, memory, packaging, equipment, materials, servers, racks, or factories, bring us the thing that is hard to qualify, integrate, manufacture, or scale.
A research proposal needs a question, a lawful input, an intended learning result, rights, funding, a qualified owner, and a stop decision. A customer product or qualification request can enter through the hardware and Foundry page instead.