Qolab’s and Dr John Martinis’s mission to industrialize quantum computing

During the National Quantum Federated Foundry (NQFF) Industry Day in Singapore, Dr John Martinis delivered a lecture on the History of Superconducting Qubits, covering his work from his Nobel Prize-winning experiment to today. 

The first part of the lecture has been published here. Below is a condensed transcript of the second half, where Martinis discusses the challenges in manufacturing scalable quantum computers, issues in fabrication, and the mission of his new company Qolab. Appended to the transcript is a short interview with Martinis and Qolab CEO & Co-Founder Alan Ho. 

Building scalable quantum computers through advanced semiconductor integration

Since leaving Google, our central objective has been to define a credible and scalable path toward building a much larger and substantially more capable quantum computer within the next several years. The challenge we face today can be clearly understood by examining the architecture of current quantum systems, including those demonstrated by Google in 2019 and those being pursued across the industry.

Most existing platforms rely on a small quantum chip—typically one to two centimeters across—with a dense collection of wires attached to it. From a physics and prototyping standpoint, this approach is entirely reasonable.

However, when viewed from a systems and manufacturing perspective, this architecture reminds you of what classical computers of the 1950s and 1960s looked like, where a small number of active elements were connected by a large number of external wires. While such systems functioned, the historical transition that enabled modern classical computing occurred in the 1970s with the advent of the integrated microprocessor, where transistors and interconnects were fabricated together on a single chip. This integration dramatically reduced cost, improved performance, and enabled scaling. Our ambition in quantum computing is to achieve a similar transformation by integrating qubits, wiring, and control elements into manufacturable, wafer-scale systems.

Advancing qubit fabrication using industrial semiconductor tools

One of the big things we need to do is to improve qubit fabrication. At present, much of quantum hardware fabrication remains artisanal in nature, often performed in academic or semi-industrial cleanrooms. We need to do very much more sophisticated fabrication.

For example, we are working with Applied Materials which provides access to custom deposition tools, typically valued between $100 million and $200 million, certainly beyond the price scale of what we are doing as a small company, and beyond what even large quantum companies could justify purchasing outright.

By collaborating directly with the tool manufacturer, we can utilize these advanced systems within their R&D cleanroom and use custom processes that can be configured specifically for qubit fabrication. This arrangement allows us to prototype qubits using state-of-the-art industrial processes, refine them through iterative experimentation, and over time if we need more volume, Applied Materials knows how to put this into a dedicated line. That’s a nice business model.

A major limitation of current quantum fabrication is the widespread use of lift-off technology, which date back to the 1950s and 1960s. Although lift-off works pretty well for academic research, it’s not what you want to do in the long run.

In contrast, we have developed an advanced deposition-and-etch system using SiO₂ scaffolding that gets developed away, allowing the formation of submicron tunnel junctions, which we think is a much more reliable way to build these devices. With Applied Materials we can have very clean interfaces, including aluminum-on-silicon interfaces that are in-situ clean, with micron-scale single-crystal aluminum. I haven’t seen anything nearly as beautiful in the industry.

Current systems rely on little microwave components and bulky wiring assemblies. Our approach is to integrate these elements directly onto silicon wafers.

Integrated cryogenic filtering

One of the first concrete projects in this integration roadmap focuses on filtering. In current systems, each line includes a microwave filter—roughly the size of a short pencil—designed to filter high-temperature noise before it reaches the qubit.

At present, these filters are typically made using paramagnetic materials such as Eccosorb, which were designed for room-temperature operations. At cryogenic temperatures, their impedance is poorly controlled, they have variable reflections, and there is concern that they may exhibit ferromagnetic behavior at low temperatures.

Our solution is to replace these discrete absorptive filters with microfabricated coplanar low-pass filters. These filters can be precisely designed using tools such as HFSS, achieving controlled impedance, tailored attenuation, and return losses on the order of 20 dB. And when you cool it down, there’s no unusual physics to deal with.

The first project we want to work on is to put that on an integrated circuit. We’ve done that as a prototype at our lab in Wisconsin, but we will use the fabrication facilities in Singapore to go beyond this test device and make it more compact.

We have already fabricated and tested prototype devices, and the results indicate improved performance, greater reliability, and significantly reduced size. The next step is to scale these designs to full wafers and develop robust manufacturing processes, potentially in collaboration with national laboratories and external partners.

Ultimately, our goal is to assemble these components—consisting of stacked dilution refrigerators—that will connect many chips together to build millions of qubits, which will be connected to racks of electronics. These electronics will handle control, readout, and error-correction.

The heart of this system remains the integrated circuits and packaging within the refrigerator—that’s what we are getting started on with A*STAR, and we are very excited about the possibilities of working together.

Quantum computing as a coprocessor

Looking forward, quantum computers are not expected to replace classical computers. Instead, they will function as specialized coprocessors, tightly integrated with high-performance classical systems.

In our collaborations, including work with Hewlett Packard Enterprise, we envision quantum processors operating alongside CPUs and GPUs within supercomputing environments.

[Alan Ho: You may be surprised to learn that the integration of quantum computing and artificial intelligence is already underway, particularly in fields like drug discovery. For example, when using generative AI to design new molecules, researchers can instruct a foundation model to create molecules with specific properties—say properties A, B, C, and D. These properties are determined using quantum mechanical calculations.

Currently, the generative AI system might produce a set of 50,000 candidate molecules. However, each molecule’s properties need to be validated to ensure they meet the specified criteria. Today, this verification is performed by sending the generated molecules to a classical supercomputer, which then carries out quantum calculations to confirm their properties. This computational process narrows down the initial set to a much smaller group, such as 5,000 molecules, that actually satisfy the desired characteristics.

Looking ahead, quantum computers are expected to enhance this workflow by offering even more accurate and efficient filtering. Since quantum computers are capable of performing quantum mechanical calculations with greater precision, they will be inserted into this pipeline to further refine the selection of molecules generated by AI. Even today, quantum calculations are an integral part of these processes, though they are typically executed on classical computers.]

Platform choice and the role of superconducting qubits

Quantum computing platforms are highly competitive, and this competition benefits the entire field. Neutral atoms and ion traps have made significant progress, particularly in scalability. However, superconducting qubits offer key long-term system-engineering advantages. Notably, gate speeds for superconducting qubits are on the order of 20–50 nanoseconds, compared to 1,000x or 10,000x slower for other platforms.

Equally important is manufacturability. Superconducting and other solid-state qubits can leverage the existing semiconductor ecosystem, enabling cost-effective, high-quality fabrication at scale. While quantum computers will remain expensive, this approach offers the clearest path to reducing cost over time.

After decades of research, we now understand the physics of superconducting qubits sufficiently well. While there is still room for refinement, the dominant challenges are no longer fundamental physics, but manufacturing, integration, and system engineering.

By adopting a horizontally integrated, semiconductor-driven development model, we can bring quantum computing into alignment with the proven methods that enabled the classical computing revolution. This is the path we believe will ultimately lead to practical, large-scale quantum computers.

This approach aligns closely with existing semiconductor manufacturing paradigms and positions quantum computing to follow a trajectory similar to Moore’s Law, driven by integration, yield improvement, and cost reduction rather than purely by fundamental physics advances.

[Alan Ho: If you take a look at all the variety of quantum computing technologies out there, most likely it’s the solid-state quantum computing technologies — spin qubit, superconducting and possibly topological, if that really works — that can take advantage of the low cost, high-quality fabrication techniques of the semiconductor industry.]

Interview with Dr. John Martinis and Qolab CEO & Co-founder Alan Ho

John, you’ve mentioned previously that you were hitting a wall after leaving Google, and that you had four or five new ideas at that time, which kind of continued along your enthusiasm for the research. Could you talk through what they were, in a very broad sense?

John Martinis: The four or five ideas are really the foundation of Qolab and the company we made. We wanted to make a quantum computer that’s more manufacturable. And what people are doing right now is nice, but it’s kind of a scientific instrument. We had to come up with a bunch of ideas in order to figure out how to build it in a more manufacturable way.

Some of the ideas are on making the qubits better and getting rid of some process step that’s kind of from the 1960s and not very good. The other has to do with how to thermally isolate the wafers and how to connect up all the wirings in some appropriate way.

The point was, how do we make it more manufacturable? How do you make it more reliable, better? And then it was a matter of just thinking about all the things that I wanted to improve. And a big part of it was also starting to work with Alan. Alan had some ideas on that too. Also the other co‑founder, Robert McDermott, who’s not here today, he had some ideas. It’s really putting everything together.

But it’s not just one simple thing, and we can make a quantum computer better. It’s really many things. It’s a big systems engineering problem, and you have to kind of solve all the problems at once in the appropriate way.

What do you think right now is the biggest bottleneck that you — and the industry as a whole — are facing? Is it a cost issue? Is it a talent issue? Is it a systems engineering issue?

John Martinis: We need more funding. But I would say the good news there is, we have the ideas, and we’ve been working with the semiconductor companies, for example, to develop better processes. That’s going all very well.

We have a lot of new things to do, and we have to work at it one by one to do that. So it’s a matter of raising the funds. It’s a matter of hiring the right people. It’s forming collaborations with these big semiconductor companies — also Singapore, who can do some of the manufacturing for us — and just putting all that together.

Alan Ho: The big idea for Qolab—which is short for Quantum Collaboration—is to, instead of reinventing the wheel, can we adapt the existing semiconductor manufacturing techniques to scale quantum computers? Because if you look at the semiconductor industry—one chip could have almost a trillion transistors, right? And you can do it at scale. You got hundreds of thousands of computers made per month.

How do we take all that capability and bring it to quantum computers? Today, it’s actually very modest, because the qubit counts are like 100. We know that we need to get to millions. So can we take these technologies, that’re proven to build trillions of transistors, to help us scale up to millions of qubits?

That’s kind of the big idea. And to modify all those various manufacturing techniques that are in the semiconductor industry for quantum—that’s where the four or five big ideas are coming from.

John Martinis: And it’s not trivial. The technology is there, but we have to figure out how to fit the superconducting technology and what people are already doing. And the semiconductor industry certainly has the expertise on how to do the materials, and we have the expertise on how to think about quantum, and it’s a matter of working together to do that efficiently. 

You’ve expressed that a lot of quantum computing announcements and roadmaps are mostly hype.  What about it do you think is hype?

John Martinis: I would say they tend to underestimate the systems engineering challenges, and getting many things to work at the same time. And I think part of that has to do with manufacturing and how hard it is to manufacture a computer.

Sure, right now we’re at the point where we’re building single machines and figuring out how to do that, but I think part of it is that physicists—scientists—tend to look at things differently than an engineer, or computer engineer, or business people who are trying to build a proper scalable business. So it’s just a matter of looking at it in a different way, in my view.

Alan Ho: Scientists like talking about what’s good about their solution, but engineers talk about what’s the biggest problem with their solution. A lot of this hype that comes out is just that people are promoting what’s really strong about their solution. They don’t highlight the things that are problems. So we think that the entire community can benefit by talking more about what the problems are that need to be solved. 

2³⁰⁰ is more than the number of atoms in the universe, right? So with a 300‑qubit quantum computer, could you theoretically model the entire universe?

John Martinis: No, that’s not how to look at it. I’m just mostly pointing out that the degree of quantum parallelism on machines we hope to build soon is actually quite large, and we can do that. But that doesn’t necessarily mean we can model the universe. You’d have to invent the right algorithm [for that].