This article is more than 1 year old
Intel sprinkles 12-qubit quantum test chips into the hands of researchers
When they're done, there might be a commercially viable system sometime after 2030... maybe
Intel is providing its latest quantum chips to research facilities, including some US universities, in order to drive the development of technology for quantum computers including techniques for handling multiple qubits.
The Santa Clara chipmaker is delivering the Tunnel Falls 12-qubit test chip to research laboratories for experimentation. Among the first to get access will be the the University of Maryland, the University of Rochester, and the University of Wisconsin at Madison and on the national lab front, Sandia.
"What we want to do here is build a research ecosystem," Intel's director of Quantum Hardware, James Clarke told us.
"If we take a look at all the big advances we've seen in the semiconductor community, both from a performance perspective of transistors, but also an ecosystem of maybe the tools to build the transistors, they've all gone through this space, where we need an ecosystem of people studying these devices to accelerate quantum to hit that milestone in the next decade."
The milestone Clarke is referring to is a commercially relevant quantum system, which he doesn't believe will happen "until well after 2030," but Intel is laying the groundwork now for the technologies to try to make it happen.
Tunnel Falls is a 12-qubit test chip based on Intel's silicon quantum dot technology. It is manufactured at Intel's D-1 R&D fabrication plant in Oregon as part of a standard 300mm wafer using similar processes to the company's normal chips.
This is part of Intel's approach to developing quantum technology, which is to "ride the coattails" of the semiconductor industry's decades of painstaking work to miniaturize and perfect CMOS transistors, Clarke explained.
"We have focused our quantum technology on the baseline technology of our CMOS processes, we're building our quantum chip using all the tools that we have to make our transistors," he said.
Although it sounds esoteric, the silicon quantum dot technology is actually just based on single electron transistors. A single electron gets trapped under the transistor gate, and that electron has a quantum property called spin that is either up or down, which is used to represent the zero or one of a qubit.
There are also sensor devices on the chip near the individual gates, which indicate whether the electron is in the spin up or spin down state.
As a test chip, Tunnel Falls implements a variety of different structures, representing different skews on the qubit design. All of this has to work in a refrigerator chilled to 1.6 Kelvin (-271 Celsius or -456 Fahrenheit).
Clarke claims Intel has seen a 95 percent yield rate across the wafer, plus "a voltage uniformity in how these devices turn on at a very low temperature that's actually quite similar to a CMOS logic process." And while the quantum performance is still being characterized, "the yield and uniformity is consistent with that of an advanced [CMOS] process," he claimed.
- IBM asks UChicago, UTokyo for help building a 100K qubit quantum supercomputer
- A lone Nvidia GPU speeds past the physics-straining might of a quantum computer – in these apps at least
- You don't have to wait for quantum computing to prepare for it
- Quantum computing: Hype or reality? OVH says businesses would be better off prepared
The next step now is to collaborate with the research laboratories to build an ecosystem that can work towards realizing workable quantum systems.
"Tunnel Falls is based on CMOS technology. That's how we're going to move fast. That's how we're going to turn the 42 year timeline for transistors into a much shorter timeline for qubits just piggybacking on what we know with transistors, and that sets us apart from others," Clarke said.
"This chip will enable novel experiments and the creation of new techniques for working with multiple qubits. And while we are testing this chip, just like with any other Intel process, we're taping out our next design, and we're designing the next chip beyond that.
"We will continue to improve the performance of Tunnel Falls. It happens almost weekly. And then I would say as a teaser, we expect the next generation chip to be released in 2024."
Long term, Intel hopes to be able to deliver a complete quantum system, not just the quantum chips.
"Intel will deliver the full stack," Clarke said. "A large scale quantum computer is going to have a quantum chip plus a host of control chips, and they're going to be based on Intel technologies, it's probably going to have an HPC system or supercomputer connected to the quantum computer to process the exponential amount of data.
"And so Intel is uniquely situated here with our capability, both from an architecture and fabrication, to build all the components. But do we sell systems? Or do we offer quantum-as-a-service in the cloud? I think that's too early to tell. Let's get the quantum system first."
He warned that this is still a long way off, contrasting Intel's step-by-step approach against some rival quantum companies that have tried to go more or less directly to attempting a functional quantum system.
"I was often asked how long would it take to have a quantum computer, I said 10 to 15 years," Clarke said. "Other quantum companies were saying they'd have a commercially relevant quantum computer in five years.
"Today you can find systems of up to a couple of hundred qubits. The purpose of these systems is to begin to explore contrived applications where the power could potentially exceed that of a supercomputer.
"In several years' time, perhaps four to six years, we will be in the range of having thousands of qubits and being able to perform error correction, taking thousands of physical qubits, and running an algorithm to allow us to have a logical or long-lived qubit with which we can do substantial manipulation without losing information."
But it will likely require error correction and millions of qubits to do something commercially relevant, he insisted, and that is unlikely to happen until sometime well beyond 2030. ®