extremetech.com

Google Reveals 'Willow' Quantum Computing Chip

Big tech is currently obsessed with scaling up artificial intelligence processing, but Google hasn't totally forgotten about its bet on quantum computing. The company has announced a new quantum processor known as Willow, which it claims is capable of solving a problem in five minutes that would take the fastest traditional computers 10 septillion years—that's a one followed by 25 zeros. Google says this advancement gets us closer than ever to a usable quantum computer, but you still shouldn't hold your breath waiting for that.

Quantum computing has the potential to be much faster than classical computing because it leverages the bizarre properties of matter that emerge at the smallest of scales. Quantum bits (or qubits) rely on superposition and entanglement to run some calculations orders of magnitude faster than classical computers, as seen in Google's laboratory tests of Willow. However, the processors need to be isolated from all outside influence in ultra-cold temperatures as even a tiny bit of errant energy can cause decoherence—a breakdown of programmed quantum states.

According to Google, Willow is much faster than its previous designs, but the real advancement is its accuracy. Even when a quantum computer is properly isolated and running perfectly, the qubit output is a probabilistic measurement of a classical bit. That means it's wrong sometimes. In general, quantum computers produce more errors the more qubits they have. But Google says it has reversed that trend with Willow, which has 105 qubits.

Google cryostat

Google uses devices called cryostats to keep quantum chips near absolute zero to limit information exchange with the environment. Credit: Google

Google's head of quantum computing, Hartmut Neven, says Willow is the first "below threshold" quantum system, which gets more accurate as the number of qubits increases. Google tested arrays of 3x3 qubits, scaling up to 5x5 and 7x7—each time, the error rate dropped. Error correction is also extremely fast because the qubits arrays have longer lifespans than individual qubits (i.e. the quantum coherence time). Researchers call this a "beyond breakeven" system because it's capable of real-time error correction. If you can't correct errors quickly enough, they spoil the calculation before it's done. That's not a problem for Willow.

It's how Willow can do in five minutes what would be functionally impossible for Frontier, one of the world's most powerful supercomputers. With a performance gap greater than the age of the universe, you might expect Willow to revolutionize computing any day now. However, the test Google uses is specifically designed for quantum systems. It's called the random circuit sampling (RCS) benchmark, which Google researchers devised several years ago and has since become standard in the field. The test is designed to be taxing for quantum systems, which leaves classical computers in the dust. It's mainly useful to demonstrate improvements in quantum computing, and it certainly does that for Willow. In 2019, Google said its Sycamore chip could run calculations in three minutes that would take a classical supercomputer 10,000 years to finish.

Google has published its results in the journal Nature, and it invites researchers to peruse its open-source quantum software. While most of Google is focused on classical computing for AI, quantum computing could revolutionize artificial intelligence. Neven points out that quantum algorithms scale much better and could be ideal for running enormous AI models. Quantum computing could also be key to drug discovery, climate modeling, and other problems that push the bounds of classical computing.

Google's Quantum AI lab has made steady advances over the years, but Willow is still just an experiment. The team's goal is to perform the first "useful, beyond-classical" computation with real-world utility. That could lead to quantum computers getting work done in the coming years, but Google isn't committing to a timeline.

Read full news in source page