One of the potential limiting factors to scaling up a quantum computer to thousand or millions of qubits involves the ability to control each individual qubit with high speed, high quality, and without interfering with a neighboring qubit. This becomes an even greater challenge when the qubits use a technology that requires supercooling them to millikelvin temperatures.
The current generation of superconducting quantum computers rely on large numbers of individual coaxial cables to route the control signals from the waveform generators that run at room temperatures and go down the dilution refrigerator in order to control an individual qubit. Each qubit requires one or more signals that needs to be routed. So for example, the Google Sycamore processor which has 54 qubits requires about 200 coaxial signals for the controls. It turned out that one of the 54 qubits was not usable; not because the qubit was bad, but because there was a problem with one of the coaxial cables that was connected to it. (Picture of Google Sycamore Processor, credit Google)
Now think about how such a system can be scaled up to 1000, 2000, or even more qubits. Routing thousands of microwave signals from room temperature to the qubits running at millikelvin temperatures would be an engineering nightmare. Not only would the engineers have to deal with mechanical issues, but there would be major thermal and electrical issues that would need to be resolved too.
So Intel and QuTech have been working hard to find a better way. Their eventual approach is to integrate the qubits and the control logic onto the same chip to that the lines connecting the two are very short and can be implemented with metal connection layers right on the chip. For those of you familiar with how Intel’s microprocessor designs evolved, this approach may sound vaguely familiar. Modern microprocessors all use a high speed cache memory. But at one time the microprocessor and the cache were implemented with separate chips and built with slightly different process technologies. Later on, as the process technologies advanced and the requirements for performance continued to increase, Intel was able to integrate the cache memories onto same chip and achieved both an increase in performance as well as improved cost due to the integration. So it looks like Intel has dusted off an old playbook and looking to do something similar once again.
However, there are still significant challenges to make this happen. Control electronics chips have historically run at a temperature of about 300 degrees Kelvin (27 degrees Celsius or 80 degrees Fahrenheit) while silicon spin qubits have been typically run at about 0.040 degrees Kevin and require very expensive dilution refrigerators that cool them down to those extreme levels. In order to integrate them, one needs to do additional engineering work so these two pieces can run at the same temperature somewhere between those two extremes.
The first step in this process was to design a qubit control chip that could run at a low temperature of 3 degrees Kelvin. Semiconductor physics changes at these low temperatures so adjustments need to be made to the process technology and design rules so that the transistors can stay operational at those low temperatures. Intel and QuTech took the first step in this with the development of a qubit control chip called Horse Ridge that was designed to run at the 3 Kelvin temperature.
This week they announced a second step with the development of a version of their spin qubit technology that they call “Hot Qubits” that can implement two qubit gates at a temperature of 1.1 Kelvin. Although the term “Hot” may be debatable when you are still close to absolute zero, it still is better than the temperature of 0.040 degrees Kelvin that was previously required. In addition, operating the qubits at a temperature above 1 Kelvin has a tremendous advantage with the reduction of cooling requirements. Operation at 0.040 Kelvin requires a very expensive dilution refrigerator that can cost upwards of $500,000 while temperatures of 1-3 Kevin can be supported using refrigerators that use liquid Helium that are much more efficient and are orders of magnitude cheaper.
Although there is still a small gap of 1.9 degrees between the Horse Ridge at 3 degrees and the “Hot Qubits” at 1.1 degrees, we expect with some additional engineering Intel and QuTech will figure out a way to close the gap and take the third step to create an integrated chip. It may still take a couple of years for this to happen, but it would pave the way in the future for much larger, lower cost and more capable quantum systems based upon their spin qubit technology with integrated control logic.
Intel and QuTech aren’t the only ones working on a hot qubit technology. The University of New South Wales (UNSW) and others published a paper last year that showed operation of single qubit spin qubit gates at a temperature of 1.5 Kelvin. Since the Intel/QuTech paper indicated they can achieve both single qubit and two qubit gates at the higher temperature, they can claim they are a little farther ahead. But we expect that UNSW isn’t too far behind and will also be able to demonstrate two qubit gates at the higher temperature soon.
For more, here are additional press releases and links to papers from both Intel/QuTech as well as from UNSW.
- Intel News Release
- QuTech News Release
- Intel/QuTech Paper Published in Nature Magazine
- UNSW Paper Published in Nature Magazine
- UNSW Hot Qubits Video
April 15, 2020