Nvidia speeds chip design

A graphics processing unit is a chip packed with tiny building blocks, and moving those blocks to a new factory process used to take Nvidia months of handwork. Nvidia now says one of those jobs can run overnight with artificial intelligence, though engineers still have to check the result. (tomshardware.com) Nvidia chief scientist Bill Dally said porting a standard cell library of about 2,500 to 3,000 cells to a new semiconductor process used to take eight engineers about 10 months, or 80 person-months. He said a reinforcement-learning tool called NB-Cell now does that work overnight on one graphics processing unit. (tech.yahoo.com) Dally said the automated layouts match or beat human versions on cell size, power dissipation and delay, which are three of the basic measures chip teams use to judge a design. He also said Nvidia is using artificial intelligence in other parts of chip design, including design exploration, bug handling and verification. (tech.yahoo.com) A standard cell library is the catalog of tiny logic parts a chip team reuses over and over, like a box of pre-made Lego bricks for silicon. Each time a chip company moves to a new manufacturing node, those parts have to be rebuilt and checked against the new factory’s rules before larger designs can use them. (cacm.acm.org) Nvidia has been building design-automation software for years, and its research group says its work spans chip design from register-transfer level design to verification, logic synthesis, physical design and manufacturing sign-off. That makes Dally’s example part of a broader push to use machine learning inside the company’s own engineering flow, not a one-off demo. (research.nvidia.com) Nvidia is also pairing that message with a new quantum-computing software push. On April 14, 2026, the company announced Ising, which it called the first family of open-source quantum artificial-intelligence models for calibration and error correction. (nvidianews.nvidia.com) Quantum processors are hard to keep stable, so researchers constantly retune them and try to catch errors before they spread. Nvidia said Ising Calibration can cut calibration time from days to hours, and said its decoding models run up to 2.5 times faster and 3 times more accurately than traditional approaches. (nvidianews.nvidia.com) Nvidia said adopters of Ising already include Harvard John A. Paulson School of Engineering and Applied Sciences, Fermi National Accelerator Laboratory, Lawrence Berkeley National Laboratory’s Advanced Quantum Testbed, IQM Quantum Computers and Infleqtion. Jensen Huang said in the announcement that artificial intelligence is becoming the “control plane” for quantum machines. (nvidianews.nvidia.com) The company is still drawing a line between automation and autonomy. Tom’s Hardware reported Dally said Nvidia is “a long way” from having artificial intelligence design chips without human input, even as the company pushes it into more steps of the process. (tomshardware.com) So Nvidia’s pitch is not that engineers are gone; it is that more of the repetitive groundwork can move from months to a night of compute. In both chip design and quantum hardware, the company is betting that faster iteration matters as much as faster chips. (tomshardware.com)