Everyone is well aware that Intel is the leader in producing classical microprocessors so there has been a lot of curiosity about what they are doing in quantum. They have had a research group in Hillsboro, Oregon working on this for several years and we had a chance to talk with Jim Clarke, Intel Lab’s Director of Quantum Hardware about their activity and also attended the recent APS March Meeting in Chicago where Intel presented 14 papers. Intel has been partnering with QuTech in the Netherlands on quantum computing and recently announced they will install a quantum computing test bed at Argonne National Laboratory later this year. They have also been doing some work with NIST evidenced by a few joint papers authored by the companies at the APS meeting.
Here is a summary of some of the key concepts that are driving Intel’s quantum strategy:
Intel believes that useful quantum computing is still a long way off. Perhaps a decade or so from now. They realize the engineering complexities associated with developing quantum technology and have a long-term roadmap to provide powerful quantum processor chips with error correction in the future. They are not planning on providing offerings targeted for NISQ applications.
Intel realizes that their biggest advantage in the QC race is to leverage the high-performance capabilities they have developed over the past 50 years in building silicon transistors. Although Intel did experiment a few years ago with the superconducting technology and built a 49 qubit chip called Tangle Lake they are now focused on building qubits with a spin qubit (aka quantum dot) technology. Spin qubits have a significant advantage over superconducting qubits because the die area per qubit is orders of magnitude smaller. This allows them to fit over 10,000 quantum dot arrays on a single 300 mm wafer. Using spin qubits also allows them to build the quantum dots in the same high volume fab facility where they build their microprocessors. Unlike some other groups that are also building spin qubits using electron beam lithography, atomic layer deposition, and lift-off silicon processing, Intel is using their standard EUV (Extreme Ultra Violet) optical lithography, plasma etch, and CMP (Chemical Mechanical Polishing) and building their chips using a high volume 193 nm lithography process. Because they have optimized these process steps for their transistor technology over many years in their fabrication facilities, this approach will provide them with high yields, high precision, low contamination, high uniformity, and high reproducibility. Intel indicated to us that no new fabrication equipment will need to be installed to build these wafers.
Intel has a lot of experience working with vendors to obtain the high-quality materials and chemicals and they will be leveraging this expertise to obtain a more purified silicon starting material called 28Si. This is an isotope of the silicon atom that does not exhibit any nuclear spin and is helpful to achieve improved coherence times. For classical transistor processing, they use natural silicon which contains a mixture of 28Si, 29Si, and 30Si. The latter two isotopes have an inherent nuclear spin which is why they want to avoid them in a quantum qubit. But for building a classical transistor, this is not an issue. Although some of Intel’s early spin qubit results used the natural silicon, they will be converting to the purified 28Si in the future.
The devices that Intel mentioned at the APS March Meeting were linear configured devices with what Intel called gate sizes of 55, 23, 17, and 7. This nomenclature is different from what is normally called a quantum gate because it represents the number of Accumulation, Plunger, and Barrier structures in the design which control where an electron can be trapped. (See diagram below.) These configurations only support nearest neighbor coupling with the smallest configuration providing 3 qubits and the largest supporting a few dozen. Interestingly, the technical results they reported were based upon qubits built with natural silicon but they still reported good coherence and fidelity numbers with T1 at about 14-65 milliseconds and T2* about 1 microseconds. Single-qubit randomized benchmarking fidelities approached 99.9%. These measurements will undoubtedly improve once they build new devices with the purified 28Si.
Although Intel won’t require new equipment to build the chips, they have been working with BlueFors and Afore to create a new piece of test equipment to probe the wafers. They have worked with these companies to create a first of a kind cryoprober that can probe a wafer at a cryogenic temperature of 1.6 kelvin in a very short period of time. One of the special challenges for teams developing superconducting or spin qubit chips is the fact that the chips need to be cooled to kelvin or millikelvin temperatures before they can be tested. Normal transistors do not have this problem because their wafers can be tested at room temperature to provide engineers with quick feedback on the results of a newly fabricated wafer. The cryoprober allows Intel to take a new wafer stick in the chamber and start getting results within a couple of hours. This allows them to rapidly assess how well the processing went for feedback to the process engineers, figure out which dice on the wafer are the best (called the Hero Devices) and then package those best chips and install them inside a dilution refrigerator for further testing. Without a cryoprober testing a chip requires sticking the device into a dilution refrigerator and waiting a couple of days before the refrigerator cools down to the millikelvin temperatures so an engineer can start testing the chip. Intel says that this cryoprober capability provides a 1000X improvement and is a major advantage for Intel’s engineers to allow them to iterate and improve their designs much faster.
Another key technology that Intel is working on is to develop a cryoCMOS control chip that can be placed close to the qubits and eliminating much of the cabling that would otherwise be required to snake through the dilution refrigerator from the room temperature electronics to the qubit chip. This can be a major mechanical engineering concern, particularly as the number of qubits continues to scale towards the 1000 or more per system. Intel has developed such a chip called Horseridge II with QuTech which they are currently testing.
One interesting point about Intel’s quantum research developments is the fact that they are not stopping at just developing the hardware. They are pursuing a full-stack approach including software as shown in the diagram below:
They have developed their own Software Development Kit (SDK) with an LLVM-based C++ compiler and system software workflow that is designed for efficient execution of classical/quantum variational algorithms. It will include its own optimizing compiler that will take a user’s program and compile it to use the processors native gate set in the most efficient manner and control all the interactions between the classical processor and the quantum processor so they can efficiently work together. The SDK will support as backends a couple of different simulators they have developed as well as a quantum dot chip. Intel also has a software team researching algorithm to understand how these could be run on a spin qubit based quantum processor.
As we indicated at the beginning of this article, Intel views quantum technology development as still being in a very early stage, and they are not planning on offering end user access over the cloud anytime soon. However, they did announce they will be partnering with Argonne National Laboratory and providing Argonne with a quantum test bed later this year. We expect this test bed to have a small number of qubits and its main purpose will be to provide additional testing and feedback on Intel’s spin qubit technology. It’s not clear that the configuration of this machine has been finalized yet, but it was not specified in the press release.
Longer team, it doesn’t appear that Intel’s ultimate business model for providing commercial quantum products have been fully determined yet. They have a number of options including selling individual chips, providing full quantum computers to OEMs, partnering with a cloud provider, or even becoming a cloud provider themselves. But there is still a lot of technical development work that needs to be done and the quantum ecosystem will surely change dramatically over the next few years. So, it may be best for them to keep their options open in the near term. In any event, we wish them the best of success.
For more information about Intel’s recent quantum activities, you can view a recent paper publish in Nature magazine that describes their latest advances in spin qubit technology here and you can register and access the recent Intel presentations at the APS March meeting here.
April 23, 2022