On the Road to Room Temperature Quantum Computation
How to use integrated photonics to build quantum computers that don’t require super-cooling, allowing them to be miniaturized and deployed on a mass scale.
By Matthew Collins and Zachary Vernon
Search for “quantum computer” on Google Images, and the results will be filled with pictures of shiny chandelier-like contraptions wired with spider webs of gold-coloured circuitry. These images all typically show the insides of helium dilution refrigerators fitted with the components needed to host quantum computer processors, specifically those based on superconducting circuits. Their purpose is to cool these processors down to temperatures of around 10 millikelvin — colder than the coldest places in interstellar space. The quantum processors themselves occupy only a tiny fraction of the room at the bottom of these refrigerators. Yet photos of the surrounding electrical wiring and cooling technology itself have become the stock images used to represent quantum computing.
Depending on whom you ask, having to operate at extreme temperatures can be either a bug or a feature. Experimental physicists, and the reporters tasked with writing about their progress, often highlight the need for ultra-cold temperatures in these systems to emphasize the bleeding-edge nature of their work. Quantum computers are about as advanced as commercially available hardware gets, so emphasizing the extreme operating conditions they require can help those outside the field grasp just how difficult an endeavour building a quantum computer is. Big companies pursuing quantum computing sometimes seem to boast about super-cooling, perhaps because it hammers home the message that their R&D departments aren’t afraid to tackle the toughest technological frontiers. Viewed from this non-technical lens, dilution fridges make for good PR — they can be a positive feature.
On the other hand, to engineering teams building quantum computers and most pragmatic-minded venture capitalists placing multi-million dollar bets on quantum computing hardware companies, dilution fridges are most definitely a bug. The need for expensive and bulky scientific equipment to house a company’s product constrains its deployment, slows development on the core processor technology, and limits scaling of the hardware. It is hard enough to imagine dilution fridges in cloud computing data centers, let alone people’s homes, cars, or personal devices. Efforts to rapidly prototype chips are impeded by the several days of time required to bring the processor up to room temperature, swap the chip, and then cool back down to 10 millikelvin. And engineers tasked with building larger processors with more qubits must first contend with the complexities of cramming the necessary wiring into the tight confines of a cryostat enclosure. Though not usually shown in promotional material for quantum computers, the dilution fridges used by modern superconducting quantum computers also require large, noisy and power-hungry compressors and gas handling units. The full systems are difficult to mass-manufacture, and, since they rely on mechanical pumps and pressurized gases, they will never exist as a fully solid-state technology.
At Xanadu we are taking a different approach to building quantum computers. Rather than relying on superconducting circuits, we use photonics to encode and process quantum information. A key advantage to this approach is its compatibility with room temperature operation. Currently the entire Xanadu hardware system, with the exception of one important component, operates at room temperature. Indeed, the quantum state preparation and gate sequences all take place on a single chip sitting at around 25 degrees Celsius. Because the photons we use have comparatively high energy (around 0.8 eV, some 20,000 times larger than the corresponding energy in superconducting circuits), the chances of a stray unwanted photon of thermal origin creeping into our photonic circuits and disturbing the computation is vanishingly low, even at warm ambient temperatures like those inside a lab or data centre. This allows us to very rapidly test our chips, swapping them into and out of a test station in minutes, and it means we can operate our processors within a standard server rack. This makes it much easier to imagine deploying our systems in many more contexts than can be contemplated for other approaches. It also lets us push the envelope of miniaturization even further, to the point where incorporating quantum chips into personal computers becomes conceivable. This won’t happen tomorrow, as more development is first needed. But such development has already started, and making progress to this goal is a major objective of the Xanadu hardware team. Because the photonic platform, unlike other approaches, suffers from no fundamental barriers to miniaturization, we can confidently pursue such development using the mature toolbox of integrated photonics.
This allows us to very rapidly test our chips, swapping them into and out of a test station in minutes, and it means we can operate our processors within a standard server rack.
There is a catch. As we mentioned, presently and in the short term, there is still one cryogenic component used in Xanadu’s quantum computing hardware stack. Those familiar with quantum photonics will recognize that we make heavy use of specialized photon counting detectors in our hardware stack. These photon counters — in our case, transition edge sensor (TES) detectors — read out the quantum states of the entangled beams of light after they have exited the chip, and comprise a very important part of our technology. Unfortunately, they also rely on superconductivity to do their job, and therefore require cryogenic temperatures. So the claim that “photonic quantum computers operate at room temperature” needs qualification: the chip itself sits at room temperature, but some cooling is currently needed for the detectors.
However, the role of cooling in photonic quantum computers is very different from that in other approaches. For Xanadu, the need for cryogenics is a side effect of the limitations of present-day photon detector technology: we need specialized high-performance detectors, those detectors currently need superconductivity, and superconductivity needs very cold temperatures (in this case, somewhat warmer than 10 millikelvin — TES detectors can run at 50–100 millikelvin and each generate only a few nanowatts of heat). For the types of quantum computers built by IBM and Google, for example, the cooling requirement is more fundamental. Because the electronic circuits that encode their qubits operate with frequencies in the microwave range, which at room temperature is awash with thermal photons that overwhelm the subtly encoded quantum bits with random noise, cryogenic cooling of the processor itself is absolutely vital. Furthermore, the cooling power requirements are fairly demanding, since superconducting processors dissipate significant amounts of heat. Whereas the superconducting qubit approach will always require an ultracold environment, photonics maintains viable prospects for fully room temperature operation: all we need to do is replace our detectors with ones that do not rely on superconducting components.
For the types of quantum computers built by IBM and Google, for example, the cooling requirement is more fundamental.
So what exactly are these prospects, and just how viable are they? There are a few different paths that Xanadu is exploring.
Single-photon avalanche diodes
Not all photon counting detectors require superconductivity. The classic example of a detector which is sensitive to individual photons while also working at room temperature is the single-photon avalanche diode (SPAD). These are similar to ordinary photodiodes, consisting of semiconductor junctions with a facet exposed to the beam of light to be measured. To achieve single photon sensitivity, they are electrically wired and driven in a special way that results in the absorption of individual photons triggering a burst of electrical current, similar to how a Geiger counter detects radioactive particles. These devices usually come as small, palm-sized modules from specialized optics companies like ID Quantique, and can even be chip-integrated.
The drawback of SPADs is that they work best near the visible range of wavelengths, whereas our chips operate in the telecom “C band”, which is infrared. We chose this wavelength to remain compatible with commercial telecommunication technology and networks, and also to stay within the wavelength region where silicon, one of the most common and best materials for nanophotonic waveguides, is transparent. (Some SPADs are able to detect C band photons, but they are bulkier, still require some cooling, and perform poorly.) Individual SPADs also are incapable of providing photon number resolution: they can distinguish between 0-photons and 1-or-more photons, but can’t accurately report “N photons detected” with any more information about N.
Both photon number resolution and C-band operation are important features that we would like to avoid sacrificing. Luckily, there are some promising strategies to incorporate room-temperature SPADs into our hardware stack while retaining those features. Quantum frequency conversion (QFC) can be used to change the wavelength of our photons right before they are detected, while preserving the key quantum properties that encode information. QFC has been demonstrated in a number of different ways, with one of the most promising using a nonlinear effect called four-wave mixing. In a lucky coincidence, this effect is similar to the one we already use to generate squeezed light on our chips. In fact, QFC works especially well in silicon nitride ring resonators, which are precisely the devices we use as squeezed light sources! Crucially, QFC functions just fine at room temperature, and so incorporating a QFC stage as an extra module in our quantum processors is a relatively straightforward engineering path to follow.
To provide photon number resolution, an array of many small SPADs can be fabricated, and the light from each chip output split into many separate beams, each of which illuminates a single SPAD. Provided there are lots of SPADs in the array — many more than the number of photons in the corresponding chip output channel — photon number resolution can be restored, since it becomes very unlikely that more than one photon will land on any individual SPAD. This strategy of multiplexing has been demonstrated by a number of different research groups, and is the typical way for quantum photonic devices to achieve photon number resolution when TES detectors are unavailable.
Combining QFC with multiplexing provides a viable prospect for fully room-temperature operation, but there are still a lot of hurdles to overcome. Conventional SPADs don’t quite achieve the level of performance we would like, even at shorter wavelengths where they tend to work better. This is especially true of chip-integrated versions, which have not yet enjoyed intensive development. And demonstrations of QFC in integrated devices still need optimization to increase conversion efficiency and decrease noise from unwanted parasitic nonlinear processes. Xanadu will continue to push performance on both these functionalities in the near future.
Non-Gaussian inputs & homodyne detection
At a very high level, the reason we use photon counting (either from SPADs, TES detectors, or other detectors that provide single-photon-level sensitivity) in our hardware stack is to include a “non-Gaussian element” within the class of quantum operations available to users. Without such an element, the quantum circuits that users could load and execute on our chips would all be efficiently simulable by a classical computer. To access interesting problems, a non-Gaussian element is vital.
But photon counting is not the only non-Gaussian element out there. Rather than including a non-Gaussian readout operation, it would suffice to provide users with non-Gaussian input states or inline non-Gaussian gates on-chip. Right now the inputs to our quantum circuits are squeezed vacuum states, which are Gaussian, and our gates are composed of linear optical transformations — also Gaussian. If we could synthesize a more exotic, non-Gaussian input state, or implement a non-Gaussian gate, then we would be able to eliminate photon counting from our hardware stack, along with all the cryogenic equipment needed to support it. Instead of TES detectors, we could use ordinary photodiodes to implement homodyne detection, which measures the amplitude or phase of quantum light instead of directly counting photons. Homodyne detection is a very mature technology, fully room-temperature compatible, and straightforward to integrate on-chip with high performance, so this route of “pushing the non-Gaussianity to the inputs” is especially appealing.
The challenge with this route is that non-Gaussian states and gates are notoriously difficult to generate and implement deterministically and without relying on photon counting themselves. This is due to photonics’ toughest feature: it is hard to make quantum states of light interact directly and to a sufficiently nonlinear degree with one another. Indeed, some view this feature as photonic quantum computing’s achilles heel (although it is also one of photonics’ greatest strengths: the lack of strong quantum-level nonlinearity is precisely what makes light such a fantastic carrier of quantum information).
Recent advances at the cutting edge of integrated photonics have challenged this conventionally held perspective. As photonic nanofabrication capabilities have advanced, new and extremely high-quality optical resonators have emerged as a viable way to produce non-Gaussian states or implement non-Gaussian gates deterministically, at room temperature, and without relying on photon counting. By tightly confining light inside specially engineered structures, the effective strength of the nonlinearity experienced by the optical fields can be cranked up by many orders of magnitude. Running the numbers using parameters that correspond to the most cutting-edge fabrication capabilities, one finds that the state of the art is just about good enough to start seriously trying to design sources of non-Gaussian states. Excitingly, these techniques are perfectly suited to our favourite optical wavelengths in the C-band. These promising signs have motivated the Xanadu team to push the frontiers of nanophotonic resonators for generating and implementing non-Gaussian states and gates.
Invisible cryogenics
Even without room-temperature detectors, photonic quantum computers are much less demanding in their cooling requirements than their superconducting counterparts. Photon detectors can tolerate warmer temperatures, have more modest cooling power requirements, and do not need nearly as much circuitry in the coldest parts of their cryostats. This enables Xanadu to pursue, in parallel with our efforts toward true room-temperature operation, very compact and power-efficient cooling systems for our detectors. Even before full room-temperature operation is achieved, these will one day allow Xanadu’s full computing systems to be entirely self-contained — including the full detection system — within a standard server rack enclosure about the size of a washer-dryer, relying on only a standard wall plug for power.
There are several well-tested technologies available for reaching temperatures below 100 millikelvin, as required for our specialized photon number resolving detectors. Over the last two decades innovation in cryogenics has been mostly centred on scaling to larger cold volumes with high cooling powers. This sledgehammer-like approach has been partially driven by a huge uptake of the well-known dilution refrigerator by quantum computing architectures requiring their processor chips to be operated in the millikelvin regime. Dilution fridges routinely reach a few millikelvin by mixing two isotopes of helium: ordinary helium-4, and the very rare helium-3. The major benefits of this technique are the relatively large cooling power it offers, and lack of mechanical moving parts. However dilution fridges suffer from the drawbacks of requiring expensive and tightly controlled helium-3, as well as a number of large compressors and pumps to force the helium mix around the dilution unit.
Another path to millikelvin temperatures is the adiabatic demagnetisation refrigerator (ADR). This cooling mechanism uses the unique properties of paramagnetic salts for the final stage of cooling. The molecules of a paramagnetic salt behave like small compass needles: when a magnetic field is applied to the salt these molecules and their magnetic moments give up heat energy as they align to the external field. Once the magnetic field is at full strength, the salt is cooled to equilibrium with the primary cryo cooler, and the magnetic field can be slowly reduced, with the salt absorbing heat as the magnetic moment of each molecule is knocked back out of alignment. The major advantage of ADRs is that they do not require helium-3 or bulky gas handling systems.
For photon detection applications, the amount of heat generated requires a much smaller system than the sledgehammer-like solutions offered by present-day dilution fridges and ADRs. Instead, progress can be made towards so-called “invisible cryogenics,” to produce miniature application-specific cryostats that fit seamlessly into a standard server rack. Both dilution fridge and ADR technologies have no fundamental barrier to being shrunk down to the rack scale, provided the cooling power and cold volume requirements are modest, as is the case for photon detection applications. This has already taken place for the cryostats used for operating conventional single photon detectors (those without photon number resolution), where tabletop sized systems are already commercially available for reaching as low as 800 millikelvin from companies like Photon Spot. With a number of application areas for these systems (both for photonic quantum computing and for other fields like quantum-secure networking) growing quickly, we can expect to see invisible cryogenics providing sub-100 millikelvin temperatures become available within the next couple of years. At that point, even without room-temperature operation, the full stack of the Xanadu cloud system will be fully deployable in a single commercial server rack with no requirements other than standard wall power.
The scenic route
The quantum computing community is divided on room-temperature operation. Some argue that it is both impractical and unnecessary. The reasoning goes: quantum computers are so specialized in their scope that there will never be a demand for their widespread physical deployment, and so there is no point in pushing our ability to miniaturize and mass manufacture them. Others admit that room-temperature operation would be a game-changer for how quantum computers are ultimately used, but are resigned to approaches that will forever require cryogenics.
Xanadu is more bullish on quantum computers’ role in the future technology landscape. For now, and for the next couple of years, we agree that all quantum computers will exist as complex and large machines in specialized labs and cloud centres. But just as photonics is unique in the applications it unlocks, it is the only approach that gives us a real hope of eventually building quantum computers that are as benign in their operating requirements as a standard server computer. One day, a quantum computer may even fit into a machine like the laptop on which you are reading this article. Getting to that point will be a considerable journey off the beaten track, but also a scenic one, and the destination makes it more than worthwhile.
