Berkeley's Lamborghini-Lovin' Dragon
Speed. Reliability. Style.
A collaboration of researchers from the Accelerator Technology & Applied Physics (ATAP) Division at Lawrence Berkeley National Laboratory (Berkeley Lab), the University of California, Berkeley, and the University of Massachusetts Amherst recently announced the development of QubiCML, which, for the first time, uses field-programmable gate arrays (FPGAs) that leverage machine learning (ML) to execute mid-circuit measurements on superconducting qubits. The announcement left out a very interesting detail, however, which is that the FPGAs are named after Lamborghinis. You already know that Lamborghinis are fast, and that tells you up front much of what you need to know about this technology.
Floor it!
The first time I saw a Lamborghini, it was driving the speed limit on Interstate Route 80 in New Jersey. Literally every other driver, including the driver of the car I was in, saw this as a once-in-a-lifetime opportunity to pass a Lamborghini on an open road. If anyone was going to get a speeding ticket that day, everyone knew who it would be. Ironically, therefore, you don’t always want to push a Lamborghini to its limits.
Superconducting quantum computers are a different story, though. Qubits can only hold quantum information for very short periods of time. You are in a race to get done what you need to get done, in the limited timeframe you have to do it. And you can’t risk a breakdown while you’re doing it, either, so you need the Lamborghini of control systems. It has to be both fast and reliable.
Grab the Wheel
This is not driving instruction but, yes, you normally want to grab the wheel before you floor it. The key takeaway, however, is associating the speed and reliability of Lamborghinis with these FPGA control systems. Otherwise, let’s face it, I don’t have a Lamborghini-driving story for you. The only other time I ever saw one was coming out of a restaurant in Indianapolis, so let’s jump right back into qubit control.
This new approach builds upon QubiC, Berkeley Lab’s open-source FPGA-based control system. Think about the boxes you might have seen from Quantum Machines and Qblox, but remove the boxes for a sleeker, sexier look… like a Lamborghini. By reliably executing operations faster, QubiCML potentially reduces the time the qubits need to maintain coherence, enables real-time quantum error correction (QEC), and simplifies quantum circuits by replacing some quantum operations with classical logic. QubiCML is applicable to bosonic qubits and has been successfully tested on qutrits. It currently sits outside the dilution refrigerator.
Stylin’ and Profilin’
Lamborghinis aren’t just for driving. You may have noticed that they’re often decorative, whether that’s in marketing campaigns or in cinema. Simply showing a Lamborghini projects a certain lifestyle.
Similarly, this research isn’t just for qubit control. It may not be as glamorous as Hollywood, but this work is applicable to improving the performance of particle accelerators and lasers.
Autonomous Driving
I’m not sure there’s a point to NOT driving a Lamborghini if you have one. Nevertheless, QubiCML’s operations are happening at nanosecond timescales. Therefore, it’s not just fast and reliable like a Lamborghini, it’s more analogous to a self-driving Lamborghini. So, while you may or may not want the former, we definitely need the latter.
Custom Builds
Lamborghinis are upgradable and customizable, because if you have that kind of money, there’s a difference between making a statement and making a statement. When we’re talking about the exploration of quantum systems, however, the design affects the functionality. FPGAs have limited resources and require sacrifices to be made; therefore, modularity can enhance this exploration. In other words, the FPGAs are upgradable and customizable, too.
Sticker Shock
Back in the 1980s, you could obtain a Lamborghini Countach fiberglass body kit for a chassis-lengthened Pontiac Fiero. You could look like you were driving a Lamborghini, but at a discount. Why would anyone do such a thing? In 1986, the Countach started around $250,000, whereas a Pontiac Fiero could be found for under $10,000. Lamborghinis are expensive, but you already know that.
Fortunately, Berkeley Lab receives government funding. It open-sources everything, gives it away, and encourages the community. The Lamborghini analogy might seem to be slipping here, because QubiC is free, but free is not necessarily inexpensive. You still need expertise to use it, and that’s definitely not free. So, you’re getting the speed and reliability of a free Lamborghini, but you still have the cost of ownership to think about.
Conclusion
Most think that application-specific integrated circuits (ASICs) are the way forward, but what should go on them? It’s an open question. Power consumption is a concern. Therefore, QubiCML can be looked at not only as a vehicle for the exploration of qubits, but also for prototyping ASIC designs. FPGAs are Lamborghinis in the sense that you can buy them today for their speed and reliability, whereas ASICs are maybe the flying cars of the future, providing capabilities not yet available in today’s models.
Image generated by an AI model provided by Microsoft Copilot.