Several quantum companies are discussing a strategy for building quantum computers that leverages the trillions of dollars in investments made by semiconductor manufacturers over the past 75 years. To dig deeper, we decided to speak with two different companies, GlobalFoundries and Snowcap Compute, that are taking very different approaches to applying their semiconductor skills to quantum hardware.
GlobalFoundries
The first company is GlobalFoundries, a multinational semiconductor foundry originally created in 2009 by the spin-off of AMD’s manufacturing operations. They have major facilities in Malta, New York, Dresden, Germany, and other locations.
GlobalFoundries is collaborating with several quantum hardware companies, including PsiQuantum, Equal1, Quantum Motion, Diraq, Archer, Xanadu, and perhaps others. Their initial work in quantum technology dates back to 2018 when they collaborated with the University of Toronto.
Spin qubits are the most similar to standard semiconductors, and this modality is used by Equal1, Quantum Motion, and Diraq. For spin qubits, GlobalFoundries can utilize their 22FDX process, which is a 22-nanometer, fully-depleted silicon-on-insulator (FD-SOI). GlobalFoundries does not require any new equipment to fabricate spin qubits for these companies, though these designs can necessitate unique structures that would violate normal semiconductor chip design rules. They are also considering using isotopically pure Silicon-28 starting wafers to improve qubit coherence time, which would not change the equipment or process flow.
One significant challenge with building CMOS chips that operate at low temperatures is that the process models used to simulate a design are not characterized for very low temperatures. A CMOS transistor can have vastly different electrical characteristics when operating at 15 millikelvin compared to the more typical 0-70 degrees Celsius (273-343 Kelvin) specified for many semiconductors. This makes modeling and simulating devices very difficult. The advantage of the 22FDX process is its backside connection, which allows a user to control the VT (transistor threshold voltage). Therefore, even if a company cannot model the characteristics at low temperatures, they can adjust the voltage applied to this backside connection until they achieve the appropriate transistor characteristics. Besides fabricating spin qubits, this process can also be used for cryogenic readout chips in quantum computers. The 22FDX process is fabricated in GlobalFoundries’ fab in Dresden, Germany.
The photonic work they are doing with PsiQuantum is quite different. This work requires dedicated resources that GlobalFoundries has installed in their fab in Malta, New York , requiring an investment of multiple millions of dollars. In April 2022, the U.S. Department of Defense provided $25 million to support a GlobalFoundries/PsiQuantum partnership. GlobalFoundries has collaborated very closely with PsiQuantum to develop a custom process for their needs, which has also yielded other benefits. Working with PsiQuantum has enabled GlobalFoundries to enhance their photonic processes, which can then be used to serve other customers.
A question that often arises when a world-class semiconductor foundry considers working with quantum technology companies is the expected volume. These manufacturing facilities require billions of dollars in capital investment and substantial expenses in engineering support. To be profitable, they need projects that require thousands or millions of wafers to achieve economies of scale. Although quantum projects are not at this level currently, GlobalFoundries believes that quantum technology is very important and will require high volumes in the future. This is why they are making the investment now to take a leading position in providing manufacturing services to the quantum industry as that industry grows.
Snowcap Compute
The second company we spoke to is a startup called Snowcap Compute, which is taking a much different approach that could potentially solve a problem quantum hardware companies will encounter as they scale up their systems.
It is well-known that exponential growth in large data centers is leading to exponential growth in electric power requirements. A large data center might have as many as 10,000 to 100,000 servers and consume up to 100 megawatts of power for microprocessors, GPUs, air conditioning, and other needs. This is a major concern for both power companies and data center operators, with some companies even considering co-locating a nuclear power plant next to the data center to supply power.
Snowcap, which recently received $23 million in seed venture funding, is promoting a different solution. Rather than providing CMOS-based semiconductor logic, they have decided to work with another logic family based on supercooled Josephson junctions called Single Flux Quantum (SFQ). SFQ stores information as magnetic flux quanta instead of current flowing in a transistor. Importantly, it operates at cryogenic temperatures, provides a 2-3 orders of magnitude improvement in power, and runs about 10 times faster. Some members of their team have been working with SFQ logic for over 10 years. They have a 28nm process technology, 300-millimeter wafer size fab lined up to produce these circuits, and they already have EDA support from Cadence.
Although Snowcap’s primary market for their technology is AI data centers, this technology could also be attractive to quantum hardware companies working with superconducting and spin qubit modalities that also operate at cryogenic temperatures. GQI has extensively examined the challenges quantum systems face when scaling in their report “Road to Shor Era Quantum Computing.” Two challenges to scaling stand out for superconducting and spin qubit systems: cabling and heat.
Typically, controlling the qubits in a system requires one or more wires per qubit. Most current systems generate control signals with room-temperature electronics and route their wires through the dilution refrigerator from room temperature at the top to sub-kelvin temperatures at the bottom. This approach works for now when quantum computers only have around 100 qubits, but a new approach is needed to place thousands of qubits within a single dilution refrigerator. Several companies are pursuing cryoCMOS chips for control, such as Intel, which has developed its Horse Ridge II and Pando Tree chips for this purpose. This approach solves the cabling problem, as only a small number of wires need to be routed from the room-temperature controller to the cryoCMOS chips. In the case of Pando Tree, this chip sits directly next to the qubit chip, and signals can be routed through a circuit board.
However, the heat issue still needs to be resolved. Even though the power to control an individual qubit is low, this number can increase significantly when multiplied by thousands of qubits. This power generates heat and can overwhelm the cooling capacity of the dilution refrigerator. For example, a Bluefors XLD1000 fridge has a rated cooling capacity of 30 µW at 20 mK. This will limit the number of qubits that can be placed inside a single dilution refrigerator, necessitating the large-scale quantum networking of multiple modules to achieve the size needed for a large-scale system. (A detailed analysis of this is available in the “Road to Shor Era Quantum Computing” referenced above) .
This is where SFQ logic becomes relevant. Potentially, it can significantly reduce the power needed to control qubits, allowing more qubits to be placed inside the dilution refrigerator. Additionally, the faster speed can also improve system speed, leading to shorter runtimes. A few companies are working with SFQ logic. SEEQC has been developing this technology for some time and recently announced a collaboration with IBM under DARPA’s Quantum Benchmarking Initiative. D-Wave has also been rumored to be considering SFQ logic in their gate-based quantum computer program. Snowcap is offering its services to quantum companies looking for a solution to their qubit scaling issues. While cryoCMOS may be a possible solution, using an SFQ approach might offer additional advantages.
July 29, 2025
