Shown: in simulation, a radio inside the package shrinks the hiding room for a relocated chip about fifteen-fold, and leases remove the worst failure of kill switches. Not shown: anything in silicon. The paper reports no power, area or cost. It models one device on an empty cell. The authors write: “We therefore do not claim that our numbers bound a real URLLC cell in either direction.”
Deployment depends on carriers. Public networks, the paper notes, “almost always run default 5G,” so the tight timing needs low-latency service provisioned around every data centre that hosts a controlled chip. That is a telecoms rollout, not a tape-out.
The rule that arrived this week asks for none of it. It takes effect on publication in the Federal Register, scheduled for 19 November. Markets read it as friction: TSMC’s US-listed shares closed down 3.68 percent on 17 November.
Pressure for tighter controls comes from Congress, at an unsettled moment: Nvidia’s board announced on 12 November that Jensen Huang had resigned as chief executive. The prize being guarded is large. The Lowy Institute estimates US data centres hold nearly 40 times China’s AI processing power.
Timeline: rule effective on Federal Register publication, scheduled 19 Nov 2026; comments due 60 days later; 90-day window for existing data-centre arrangements.