Dialogue with Intel Quantum computing is difficult to develop faster than Moore's Law
Intel is an active participant in quantum computing, but for a long time, Intel did not provide users with a quantum processor that could be used. At the 2022 Intel Innovation Conference in October, with the release of the beta version of the quantum SDK (software development kit), Intel joined IBM and Google to become a full-stack quantum computing company.
The Intel Quantum SDK includes an intuitive C++-based user interface, a compiler toolchain based on the underlying virtual machine (LLVM), a quantum runtime environment optimized for executing hybrid quantum classical algorithms, and a high-performance Intel Quantum Simulator (IQS) qubit target backend. Future versions will also feature different target backends, including a quantum dot qubit simulator, which will eventually provide Intel qudot qubit devices.
Recently, James Clarke, Intel's director of quantum hardware, talked about Intel's quantum strategy, and Intel said that it plays the role of a turtle in the familiar fable of the rabbit and the turtle. "I don't believe quantum computing will change the world tomorrow, and I don't believe it will never happen," Clarke said. I believe it will happen at a pace similar to innovation in the semiconductor industry. ”
James Clarke
Q: Why is Intel right and those who boast of an earlier breakthrough in quantum computing wrong?
From the history of computers, the first general-purpose computer was ENIAC. It began to be developed before the end of World War II, but did not use it until after the end of World War II; Its purpose was to calculate the trajectory of artillery. It had several thousand vacuum tubes, and it could be seen what it looked like in terms of wiring and operation, and lasted for about 10 years. It wasn't until a few years after ENIAC that the first transistors were available; I could call it microscopic, but in reality, people can see it almost with the naked eye. After that, the microelectronics industry began. But it wasn't until 1958 that there was the first integrated circuit, the first monolithic hardware.
ENIAC system
Robert Noyce is one of the inventors of integrated circuits and one of the founders of Intel. It wasn't until 13 years later that we had the Intel 4004, which is generally considered the first microprocessor with 2500 transistors on it; This was the beginning of Moore's Law. From there to a million transistors, it wasn't until around 1990 that a million transistors were realized on one chip. I compare it to quantum computing because most people agree that about a million qubits will be needed to provide a commercially viable product. For example, something that can crack (secure) code requires about a million qubits. Here, it took 43 years from the first transistor to the first million-transistor chip.
Those who think we can go from a few dozen qubits to a level with commercial quantum superiority are wrong; I think this timeframe may be comparable to the timeframe (of microelectronics). Still, given that the (quantum) community has been researching in many forms since the late '90s, it's not unreasonable for someone to think we'll achieve commercial quantum advantage within 10 to 15 years from now.
Q: Each qubit technology claims to have unique advantages such as fidelity, switching speed, scalability, etc. What's so special about Intel's qubits?
Our qubits are different from other technological routes. For example, IBM, Google, or other companies that are doing superconducting qubits, and we do spin qubits in silicon: they look a lot like a transistor. What we are doing is encoding the |0⟩ and |1⟩ of qubits into the spin of a single electron; We are essentially making single-electron transistors. We do this at Intel's largest factory, the RP1 plant and the D1 factory in Hillsboro, Oregon. We do the latest CMOS or logic chips there. Once you have qubits, you will need to control them: the state in which signals are sent in and read out. We have two ways to do this: the first is that we use Intel FPGA-based room temperature control electronics to make the control box.
Intel's silicon-based qubits
However, it is somewhat untenable to let many wires (from the control box) into the dilution chiller to control the chip. What we're going to do is place our control electronics very close to the qubit chip in the chiller. This is the Horse Ridge series of chips. We optimized our design to perform at 4 Kelvin, which is four degrees above absolute zero, or 269.15°C. Now we are developing the second generation and have some derivatives of this chip that are being tested today.
But that's not enough. Intel is not only good at manufacturing, we are also good at architecture. Therefore, we need a quantum architecture, and we have the compiler, run time, processor at the bottom of the stack. Today, we have Intel's quantum simulator. Intel was actually one of the first companies to have a quantum simulator, and it's open source. It simulates abstract qubits. At the Intel Innovation Event in early October, we launched the Intel Quantum Software Developer Kit (SDK). We are beginning to engage with developers and external partners about our software suite. Now that Intel can emulate more and better qubits, developers are doing things differently than they would in a way that has access to how the system will work. Later, we will implement a simulation of actual spin qubits. In 2023, we'll connect everything with actual spin qubit hardware.
As far as I know, rarely, perhaps only one company can do all of this in-house without having to work with other companies to implement some of it, whether it's manufacturing, control, or architecture.
Q: Is it necessary, or even the best, approach to become a full-stack quantum company?
Actually, that's a good question. I can say two things. Of course, Intel, as an IDM (Integrated Equipment Manufacturer), believes that the joint optimization of process control and architecture of traditional transistors provides performance advantages. We think the same way about quantum technology. For example, there may now be 10 different types of qubits, and perhaps 5 or 6 are talked about a lot. Imagine if a company focused solely on software, and for them to have a business model, they would have to focus equally on all the different types of qubits, or be experts in one or the other.
For me, at this stage, I think something more can be gained by optimizing the whole system.
Q: Intel will launch its own quantum processor (QPU) in 2023, can you tell us about the upcoming Intel quantum processor?
Now, we're heading toward a 12-qubit chip. 12 qubits is less than some others in the industry, but we believe our qubit system will be quite large. Spin qubits are comparable in size to transistors, meaning they are a million times smaller than superconducting qubits.
Q: How will Intel provide access to its initial 12-bit QPU? Is it through Intel's cloud, or is it using one of the existing public cloud portals?
Together with our SDKs, at the earliest stages, access will be provided through Intel's proprietary capabilities (Intel Developer Cloud). It remains to be seen how I think we will deliver it in the long run.
Q: What do you think of Intel quantum systems becoming part of HPC? Will it be very similar to the current developer community around Intel processors in the long run?
Let me divide it into parts. Large-scale quantum computers will require a high-performance computer or small supercomputer to help process data, and our vision is to bundle Intel quantum computers with Intel supercomputers. In fact, the bill of materials for the entire large system may be more on the supercomputer than in the quantum computing part. Any type of system, any type of algorithm, will be fed partly to a classical computer, partly to a quantum computer, and they will work together.
I think it's too early to know how these two systems merge. I can tell you that our team at Intel is working on these algorithms, hybrid algorithms [for] classical and quantum. The developer community will likely want to evolve and learn how to separate parts of the algorithm and send them to quantum computers.
Q: Do you think there will be a quantum model that wins? Or are there two or three?
That's a good question. I think the big companies out there, including Intel, but also IBM and Google, are currently focused on universal quantum computers. Every system has a slightly different topology—how qubits are connected and operate. But if there could be a system of a certain scale, and the system could be characterized by the number of fault-tolerant qubits, I would expect the largest system to be the best, whatever it is.
If there are ten different qubit types, I don't expect more than two to reach this stage, and I wouldn't be surprised if only one qubit type wins.
This is a big problem. Five years ago, when many companies started being more open about their quantum projects, everyone thought we would have these applications now. If you look back at the publishing media, most companies think there will be commercial quantum systems by 2022 or 2023. Intel's views are very consistent because we know how long it takes to develop a new technology. Even if we have a two-year cycle in transistor technology, the development time for these technologies is 10 to 12 years. I haven't seen any indication that there will be applications that offer huge business advantages anytime soon.
What does that mean now? Maybe there's an app that beats commercial, classic computers, but it's unlikely to be an app that makes a lot of money. And that's what we ultimately care about, something that will change the world.
Q: The speed gains discussed so far are also limited?
Yes, Shor's algorithm has a provable speed boost on quantum computers, but it has not yet been implemented. As far as I know, there is no provable speed boost for optimization problems, that is, none of them mathematically prove this advantage.
Q: What are Intel's plans for nex
We are using the unique capabilities of our quantum control chip, Horse Ridge, to characterize our 12-qubit quantum computing chip and develop a larger array.
The Horse Ridge II, Intel's second-generation cryogenic control chip, brings the key control functions of quantum computer operation to cryogenic chillers to simplify the complexity of quantum system control circuits.
Q: As you pointed out, it's still early days in terms of taking quantum computing devices out of the lab and turning them into useful devices; Do you think we'll see surprises in the development of technology in the next few years?
The only surprise is that those who think [progress] will be fast are fooling themselves. We chose a qubit that looks exactly like a transistor for a reason: we can use all the tools in our toolbox used to make transistors to make our qubits. If you pull a group of people together and ask, "Is quantum computing harder than classical computing?" "Everyone will say, yes; So why should quantum computing be thought to advance faster than Moore's Law? Don't predict the future based solely on the past, but there's no indication that it can move that fast.
In fact, the only thing I want to say is that the more we copy the tools in Moore's Law toolbox, the faster we can go. I don't think they'll actually have any surprises. I think it's going to move at a slow, steady pace, and many people who plan to make a quick profit will be disappointed and maybe even lost.
Q: Unlike the early days of integrated circuit development: the development of integrated circuits came mainly from two specific projects: the "Minuteman" and the Apollo program, and now there is a lot of money and effort from government and industry pouring into quantum computing. Given the number, diversity, and intensity of people undergoing quantum development, you still don't see an acceleration in the development process?
That's a good question. We think public/government funding for quantum could be around $24 billion around the world. That's a big number. The problem is that there are now at least five qubit types: these huge sums of money are being spread across many technologies, and there are few overlapping technologies.
The analogous transistor, microelectronics industry developed in the 90s with the "Strategic Alliance for Semiconductor Manufacturing Technology" (Sematech), and the winners were those who took these few options and made them a reality. Now, we have a whole bunch of different technologies out there with no overlap at all. So there really isn't a pre-competitive collaboration like the semiconductor industry, which also dilutes the amount of money available to accelerate the effort.
Q: The DOE is trying to promote more private and public collaboration in their quantum center and on their quantum test platform. Is Intel actually involved?
Yes, we are working on the Q-NEXT project at Argonne National Laboratory.
Finally, I would like to say that the challenge for the government is to put together a huge scientific and technological collaboration, but basically give each professor one student to work: no holistic work can be done. This will be a problem, and something we've been asking attention for: taking a huge effort, but spreading it across a lot of technology means it's unlikely to speed up anything.
