Protocol | Enterprise
Intel, Microsoft and Google’s quest to control complex quantum computers
Mo qubits, mo problems.
Photo: IBM Research
The iconic image of quantum computing is the "Google chandelier," with its hundreds of intricately arranged copper wires descending like the tendrils of a metallic jellyfish. It's a grand and impressive device, but in that tangle of wires lurks a big problem.
"If you're thinking about the long-term prospects of quantum computing, that image should be just terrifying," Jim Clarke, the director of quantum hardware at Intel, told Protocol.
The problem is scale. When your quantum computer is based on just a few dozen qubits, it's fine to designate one or more wires to each qubit in order to send and receive signals. But quantum computers are growing, and they will need to reach thousands, if not millions, of qubits to truly be useful for companies and organizations — at which point, using wires is going to become a literal knotty problem.
"It's no longer sufficient to just be able to make a qubit device and talk about the properties of a qubit," said David Reilly, a quantum researcher at Microsoft. "It's all about scale. And it's all about the infrastructure that you need to get there."
Multiple companies investing in quantum computers are now looking toward this future, and trying to develop better methods of controlling their qubits. These quantum control chips are similar to the silicon wafers produced using CMOS technology, with a few alterations to make them work at ultracold temperatures and with qubits.
In December, Intel announced its Horse Ridge 2 control chip, which builds off its predecessor to perform more specific applications, such as manipulating and reading qubits. Research soon to be published in Nature Electronics looks at Microsoft's quantum control chip, which operates at the coldest temperatures yet: a tenth of a degree above absolute zero. But Microsoft and Intel aren't alone; other companies have also been developing quantum control chips. In 2019, Google debuted its own.
The goal is to help build the infrastructure required for a future quantum computer that has actual applications for corporations and research institutes, not just one that can out-speed classical supercomputers in proof-of-principle tests.
Quantum computers rely on unsettling properties of the quantum world, such as superposition: the combination of states exemplified famously by Schrödinger's feline, which can be neither dead nor alive, but exists in some superposition of both states. Quantum effects like superposition are fragile, and tend to collapse out in our noisy, warm, macroscopic world. To preserve the qubits' delicate quantum states, they are kept extremely cold, usually below 1 Kelvin (about -457 degrees Fahrenheit)
To achieve those temperatures, qubits are kept in dilution fridges, which use liquid helium to keep their contents ultracold. Those devices have limited internal space and limited cooling power, both of which are hampered by a nest of wires. To explain the impetus for better quantum control, Clarke uses an analogy.
"Imagine that you are in charge of 100 dogs, and you have to take them for a walk," Clarke says. "Do you take them on 100 separate leashes? Or do you have one leash that fans out and do it more efficiently?"
In classical computers, control chips are small silicon wafers with some input and output pins that connect to billions of transistors. This CMOS technology is well-known, so it's tempting to slap a classical control chip onto a quantum computer, but it's not feasible. There are two main problems: First, quantum control chips have to operate at extremely low temperatures, so circuits must be designed to dissipate far less heat. Second, quantum control chips require alterations to interface with qubits, which are far more finicky than their classical counterparts.
Solving both problems entails reworking control chips: though hopefully not so much that they can't still be fabricated at scale, with traditional silicon manufacturing techniques.
Intel claims its Horse Ridge series (named for Horse Ridge, one of the coldest places in Oregon) operates at 4 K — above the temperature of the qubits it can control — and seeks to build on the company's preexisting expertise with semiconductor technology. While Horse Ridge 1 was designed for a wide variety of qubits, the recently unveiled Horse Ridge 2 focuses on expanded applications for spin qubits, Intel's preferred choice in the qubit wars. The goal, according to Clarke, is to "customize these low-temperature control chips for different applications."
Google's control chip is similar, operating at 3 K, but is designed for yet another type of qubit, called a transmon. The chip has been tested on a few qubits, but not the 53-qubit Sycamore device that made headlines in 2019. Research from Google has focused on reducing heat dissipation from the quantum control chip. Where a classical control chip outputs about one watt of heat, Google's dissipates less than two thousandths of a watt. "Getting power down is going to be very important for any of these approaches," Joseph Bardin, an engineering professor at the University of Massachusetts Amherst and a researcher at Google, wrote in an email to Protocol.
Microsoft's quantum control chip relies on similar semiconductor technology, but it is designed to work with topological qubits, though it can be adapted for other qubit architectures. (Worth noting: Topological qubits are theoretical, and have not actually been demonstrated.)
Where it really departs from Intel's and Google's designs is temperature. Microsoft's quantum control chip can dissipate less than a millionth of a watt, allowing it to stay at a chilly 0.1 K, about the same temperature as the qubits themselves.
The companies insist on the advantages of their own approaches.
"What is really unique is that our signals are generated at low temperature," Intel's Clarke said. By putting its classical CPU — which sends signals to the control chip — with the control chip at these temperatures, Intel hopes to avoid the noisiness inherent to signals generated by CPUs at high temperatures. Microsoft's Reilly disputes this claim of uniqueness, and notes that Microsoft has generated its signals at temperatures as low as 1 K.
For Reilly, running the control chip at 3 or 4 K — as Google and Intel do — is much too hot: Though the difference between 0.1 K and 4 K may sound small, it's also the difference between the quantum world and the classical world. By operating the quantum control chip at the same temperature as the qubits, Reilly hopes to avoid the bottleneck between shuffling information from quantum to classical or vice versa.
"If their control system is at 4 Kelvin, then you still have this I/O bottleneck," Reilly said. "You still have a bird's nest of cables that run from four Kelvin to the millikelvin stage where the qubits are." One possible solution to that would be to bring the qubits up to 4 K — and there is some progress in that direction.
All three companies will be presenting updates at the International Solid-State Circuits Conference, which will be held virtually Feb. 13 through Feb. 22.
For now, though, no one has remotely enough qubits to necessitate these quantum control chips, let alone determine which of those compromises is preferable. Testing is still done on a few qubits, or devices that model qubit behaviors — not even the full ensemble of dozens of qubits that make up the current cutting-edge of quantum computers.
At any rate, quantum computing is still in its heady early days, where firsts are being made, and — despite some corporate jousting about progress — demonstrating the possibility of the technology often takes precedence over long-term plans to make it practical. And at this stage, we don't even know what all those roadblocks might be.
"Wouldn't it be ironic if it turned out that making the qubits turned out to be a little bit of a challenge?" Reilly asked. "And the real roadblocks were in the systems integration and the construction of cryogenic control at scale?"