Supercomputer simulates quantum chip at scale never attempted before

Scientists just ran the most detailed physical simulation of a quantum chip ever attempted, using nearly 7,000 GPUs working in parallel for 24 hours straight. The result: a blueprint so precise it can predict how the chip will actually behave before anyone fabricates it.

The simulation happened on Perlmutter, one of the world's fastest supercomputers, housed at the National Energy Research Scientific Computing Center in California. Researchers from Lawrence Berkeley National Laboratory and UC Berkeley built a model of a quantum microchip just 10 millimeters square — smaller than your pinky nail — but so intricately detailed that capturing it required nearly the entire GPU fleet.

Here's why this matters: quantum chips are finicky. The qubits that do the actual computing are sensitive to electromagnetic interference, and even tiny design flaws can ruin their performance. Normally, engineers build the chip, test it, find problems, and redesign. That cycle is expensive and slow. This simulation skips ahead — it shows you the problems before you spend millions fabricating the hardware.

From Theory to Prediction

The team, led by researchers Zhi Jackie Yao and Andy Nonaka, used a tool called ARTEMIS to discretize the chip into 11 billion grid cells — essentially turning the physical structure into a mathematical landscape. They then ran over a million time steps to simulate how electromagnetic waves would propagate through the design, how signals would couple properly, and where unwanted crosstalk might occur.

"I'm not aware of anybody who's ever done physical modeling of microelectronic circuits at full Perlmutter system scale," Nonaka said. The scale is the breakthrough. Previous simulations made simplifying assumptions — they treated parts of the chip as "black boxes" and ignored certain physical details. This one modeled everything: the material composition, the wiring layout, the resonator geometry, the interactions between qubits.

Solving Maxwell's equations in the time domain meant the model could capture nonlinear behavior — the messy, real-world physics that matters when qubits are actually operating. The simulation was fast enough that the team tested three different circuit configurations in a single day.

The collaboration involved Irfan Siddiqi's Quantum Nanoelectronics Laboratory at UC Berkeley, which designed the chip being simulated, alongside Berkeley Lab's Advanced Quantum Testbed. That partnership — between designers, modelers, and computing infrastructure — is itself part of the story. Quantum hardware is hard enough that no single team can solve it alone.

Next comes validation. The team plans to fabricate the physical chip and compare the simulation's predictions against real measurements. If the model holds up, it becomes a tool for accelerating the entire field — every quantum chip designer could use similar simulations to iterate faster and cheaper. That's the kind of infrastructure breakthrough that compounds: it doesn't just solve one problem, it changes how problems get solved.