Work begins on building a quantum computing ecosystem.
Quantum computing is showing significant promise, and research is beginning to move from the earliest stages to a deeper understanding of what works best commercially and why.
On paper, quantum computing algorithms are potentially revolutionary. They suggest a way to solve some problems more quickly and more accurately than conventional computers ever could. But out in the real world of practical systems, creating and preserving a superposition of quantum states is a constant struggle against entropy, where the uncertainties of our macroscopic world need to be restrained from asserting themselves.
One of the biggest challenges for quantum computers is error correction, which protects calculations from degradation due to noise. The most promising approach to quantum error correction requires multiple copies of the data, with hundreds or even thousands of physical qubits for each logical qubit. Even crude quantum computers are thus likely to require millions of physical qubits.
Semiconductor manufacturing is perhaps the only mature technology that can produce millions of nearly identical nanometer-scale structures. So even though electron spin-based silicon qubits are relative newcomers — they were first demonstrated in 2012 — silicon appears prominently in lists of candidate quantum computing technologies. This is true despite research into superconducting Josephson junctions, diamond vacancies and other approaches.
The rise of silicon was particularly evident at the recent IEEE Electron Device Meeting, where Ravi Pillarisetty, senior device engineer in Intel’s Components Research group, made a forceful argument for the superior scalability of silicon qubits.
Isotopically pure silicon, 300-mm test wafers at Intel
Silicon-based qubits resemble single-electron transistors. An individual electron is trapped in a quantum dot, where it can be forced into either a positive or a negative spin state. Persistence in this state, and therefore the coherence time of the qubit, is limited by interactions with the outside world. About 92.2% of naturally occurring silicon atoms have an atomic mass of 28, with 14 protons, 14 neutrons, and no nuclear spin. About 4.7% of natural silicon, however, has an atomic mass of 29 and carries a +1/2 nuclear spin.
Interactions between carriers and nuclear spins can degrade mobility in conventional transistors. The use of isotopically pure 28Si has been proposed before, but the high cost and limited availability of the material have prevented commercial adoption. In quantum systems, though, interactions with nuclear spins can break the superposition between qubits.
Given the already daunting precision needed for quantum computation, the presence of “heavy” silicon is just too large a potential error source. For scalability, Pillarisetty said, silicon qubits need to be made with 28Si, and they need to be fabricated on 300-mm wafers with factory-grade equipment.
As the history of extreme ultraviolet lithography shows, Intel’s industry dominance gives it significant leverage with vendors. Seeing a requirement for abundant 28Si, Pillarisetty said, “Intel partnered with Urenco and Air Liquide to create an ecosystem to provide isotopically purified precursors for our 300mm epitaxial growth module at a scale that supports high volume manufacturing.”
With that limitation solved, Intel was able to confirm that isotopically pure silicon behaves as a drop-in replacement in the company’s standard CMOS process. Deposition, etch, dopant activation, etc., all behave as expected, producing CMOS transistors with a measurable mobility advantage relative to natural silicon.
The company has also produced 300-mm scale qubit test vehicles, with more than 10,000 test devices per wafer. Most quantum computation research projects are fabricating only a handful of qubits at once. A test wafer with 10,000 devices brings enough data to derive realistic performance statistics and evaluate process and design improvement strategies.
The first steps in this process flow involve traditional transistor process modules like shallow trench isolation and gate dielectric formation, and results from those test structures have been good. Full qubit structures require additional process and design steps to integrate RF transmission lines, which will couple to and manipulate individual electron spins.
The need to test quantum devices at temperatures below 2 Kelvin is a major obstacle to validation of these process steps. Testing a conventional wafer at ambient temperatures takes about an hour. Cryogenic testing of quantum devices can require as much as 12 hours per device, Pillarisetty said. Faster and more highly automated test equipment is a critical next step in the development of a quantum computing ecosystem.
Qubits and nanowires
Also at IEDM, Leti’s Maud Vinet discussed fabrication of electron and hole qubits in an SOI nanowire-like process flow. Leti’s proposed design places control gates on top of and sensing elements below the qubit array. This vertical stacking minimizes the array footprint.
Needless to say, a simpler error correction strategy would go a long way toward addressing quantum computing’s scalability concerns.
Majorana qubits may be part of the solution. At this time, Majorana qubits are mostly a theoretical construct. Attempts to demonstrate their existence in the laboratory have produced ambiguous results. In theory, though, Majorana qubits come in pairs. For example, Leo Kouwenhoven, of Microsoft Quantum Labs Delft, described a semiconducting InAs nanowire, partially covered by a superconductor.
At an appropriate electron density and magnetic field, a Majorana wave function could appear at each end of the wire. The pair would function as two halves of a single fermion, and would be protected from decoherence by the structure of the nanowire.
Conclusion
In the late 1950s, the first integrated circuit designs were just starting to emerge. It wasn’t yet clear whether silicon or germanium was a better device material. Key problems like electrical isolation and connectivity had not yet been resolved. The first single-chip microprocessor was still a decade away.
That’s about where quantum computers are today. Manufacturable qubit integration schemes are just starting to emerge. Early results are promising, the potential is clear, but we don’t know yet what the future will hold.
Update
In March 2021, the team at Microsoft Quantum Labs Delft published a significant reevaluation of their previous work on Majorana fermions, including a retraction of one of the papers discussed in this article.
Related Stories
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Hi Katherine – Great article! D-Wave should be included in the conversation, too, as they have already progressed past the R&D stage into commercialization, with sales to both NASA and Google, I believe. https://www.dwavesys.com/news/media-coverage
I’m definitely familiar with D-Wave’s work, and interviewed Eric Ladizinsky back in 2014.
https://semiengineering.com/how-to-program-a-quantum-computer/
This article, though, is specifically about work presented at the recent IEDM.
Wow! What kind of Silicon is this:
“About 92.2% of naturally occurring silicon atoms have an atomic mass of 28, with 14 protons, 14 neutrons.”