Big ticket investments have created a growing number of quantum unicorns. Chinese strides with photonic and superconducting qubits have grabbed attention. Trapped ions have shone with demonstrations of logical qubits. But who is winning the race to build a practical quantum computer?
Where should we start in assessing the most important quantum hardware developments? The challenges and opportunities faced by different qubit platforms vary markedly. Increasingly we have to understand the goals being prioritised by each company and how they interrelate. How well can each player deliver their own quantum roadmap?
For an introduction to qubit technologies read Quantum hardware – into the quantum jungle.
FTQC – The emerging standard for early fault-tolerant quantum computers is a device with 1M physical qubits running surface code (or similar) error correction to deliver up to perhaps 1000 logical qubits. To make this work, the assumption is thrown in that 2Q gate fidelity will be 99.9%+ and that qubit coherence lifetimes will be long enough that they don’t dominate error budgets. To achieve this, companies need to show that they can radically scale-up their technology. The exact challenges vary by qubit platform but never look easy: fabrication tolerances, excessive control wiring/laser alignment complexities, cross-talk/calibration, heat-load and cooling performance, quantum interconnects and the latency of classical processing for control and error correction. In most approaches the creation of ‘magic states’, a resource needed to support a universal gate set, looks like a key bottleneck.
NISQ – Some players emphasise what might be achieved sooner with more modest, noisy, intermediate scale quantum devices. These avoid the large overheads required for quantum error correction, but instead seek to complete calculations in a small number of steps (low circuit depth) so that the errors introduced by each physical qubit gate don’t become overwhelming. Fact Based Insight believes that for gate-model quantum computers to achieve broad quantum advantage in practical applications, 2Q fidelity of 99.99%+ is likely to be required. Enhanced, or even problem-specific qubit connectivity is also likely to be at a premium. Low latency integration with classical processing will be required.
FQQC – It’s also important to recall that for some applications we only require ‘a few qubits’. Early examples of such applications are often at the intersection of quantum computing, cybersecurity and quantum communications; an overlap that promises to grow ultimately into a Quantum Internet, and with sensors a Quantum Internet-of-Things. Different trade-offs here may ultimately suit different qubit platforms. Being able to offer some coherence lifetime even at higher, easier to deploy, temperatures may be a useful advantage.
Scaling up is hard, and it’s a positive for many hardware players that they can point to fidelities improving from one generation of their multi-qubit devices to the next. However, in almost all cases they are still playing ‘catch-up’ with what can be achieved with ideal 2Q lab-based devices. Frankly, the fidelities we see today are simply not good enough.
For many years, the informal entry-level benchmark for gate-model quantum computing platforms has been to demonstrate 2Q gate fidelity above 99%. With experience and a practical focus, the field now needs to realise that the minimum metric is 99.9% (or you need to provide a very clear note from your head of quantum error correction on why this bar doesn’t apply to you). As we’ll see, in 2021 companies have been taking both of these approaches.
Roadmap summaries – items in italics are ones Fact Based Insight has not seen demonstrated at the time of publication.
Superconducting qubits land a series of punches
IBM impresses with underlying roadmap progress
In 2021 IBM launched their largest processor to date, the 127Q Eagle. We don’t know yet what QV this processor will deliver, though this will certainly be limited by fidelity rather than qubit count at least in the short term. However, Eagle is already impressive. The IBM team is clearly very excited that they have managed to pull off the fabrication of a new multi-layered chip architecture with qubits, readout resonators and control lines on different layers. This significantly simplifies the control routing challenge and is a proof of concept for further scaling. Importantly, IBM also continue to demonstrate their ability to move the performance of previous generations of processors beyond earlier limits.
Fact Based Insight sees IBM’s recent advances on underlying fidelity stats as particularly important. The Falcon series of devices have now achieved a string of promising results: the R8 revision now regularly achieves average lifetimes of about 0.3ms (T1). Test devices have achieved 0.6ms . Long lifetimes are ultimately an important foundation for overall fidelity. They are also important when qubits have to wait for other operations to complete (such as during error correction cycles). Long state lifetimes have been a promise of IBM’s fixed frequency transmon qubit philosophy and so are not necessarily easy for other players to match.
In a striking 2021 announcement, the best CNOT gates on the Falcon R10 are now at 99.91% 2Q gate fidelity. The trend across previous IBM processor families is impressively consistent . In addition, IBM has published promising results with experimental devices exploring variants of the architecture such as the introduction of tuneable couplers. Small test devices have reported promising results with 99.85% 2Q gate fidelity . This technology should also help supress crosstalk in simultaneous operation.
IBM’s progress matters, because it gives us reason to believe that hitting 99.9% 2Q gate fidelity across a large-scale device with current materials and fabrication techniques is still a reasonable projection. If we have to rely on material science advances, or completely new fabrication techniques, the current aggressive timescales of the sector can only be disappointed.
IBM fabricates its own chips at its Yorktown Heights 200mm pilot line facility. It points to this being state-of-the-art in the specific context of superconducting qubits. This seems to be giving it the confidence to press on to fabricate single chips with increasing numbers of qubits. IBM also points to its extensive patent portfolio and the role it can play in protecting its fabrication techniques.
IBM have also introduced a new performance metric for early quantum devices. Where previous measures have sought to capture device size (qubit count) or quality (QV), the new measure (CLOPS) focusses on speed – how quickly an (arguably) representative sample of operations can be completed.
Qubit count – The number of available qubits (interconnected and available to take part in 2Q gates) remains a key indicator. Recall that below about 50Q we can still simulate most things that a quantum computer can do with a conventional computer.
Quantum Volume – This metric measures the largest random ‘square’ circuit that a device can faithfully implement. This combines the width in qubits with the circuit depth in QV layers. Each layer is a random permutation of qubits followed by random 2Q gates (specifically SU(4) gates) between all pairs. The physical circuit depth required is typically much larger than the number of QV layers, so the metric captures not just qubit gate fidelity, but also the benefits that flow from flexible native gate sets, enhanced qubit connectivity, effective compiler routing and low-level error mitigation.
Circuit Layer Operations Per Second – This metric measures the number of QV Layers the processor can execute per second (averaged over 100 shots). This maintains a hardware agnostic approach and allows us to capture the impact not just of fast gates, but also high-fidelity fast read-outs, low latency control systems and low-level compiler performance.
Quantum computing proponents have in the past tended to neglect raw processor speed, relying instead on the assumption that an advantage in the algorithm executed will always carry the quantum computer with the largest quantum capacity to victory. However, increasing insight into the challenges faced by real machines is forcing us to be more specific. In the short-term high repetition variational algorithms need to be executed in reasonable timeframes on NISQ devices. Moreover, current FTQC architectures look too slow to offer an advantage for algorithms that only offer a quadratic speedup. Faster gates help.
A key test of whether any benchmark is useful is does it help the community focus on real issues that need to be fixed, or does it just invite the metric to be gamed? Fact Based Insight believes that taken together, qubit count, QV and CLOPS are likely to prove useful high-level indicators. In 2019, IBM devices offered only 16 QV and 200 CLOPS (inferred); today they offer 128 QV and 2000+ CLOPS across a series of cloud accessible Falcon processors.
Eagle is the last in the line of processors making use of the IBM Quantum System One chassis. IBM have just announced the IBM Quantum System Two as the housing for future chips. This features a modular hexagonal concept encompassing control electronics and cryogenics. Delivering sufficient cooling power and minimising downtime is another scaling challenge for superconducting qubits. IBM have brought in specialist partner Bluefors to help meet this requirement. The expanded cryo capacity also leaves open the door to future coherent networking between processors using quantum interconnects.
IBM roadmap – 127Q (Eagle) 2021, 433Q (Osprey) 2022, 1121Q (Condor) 2023; leading to a 1mQ large scale system. Error correction via heavy hexagon code (hybrid surface code & Bacon Shor code). Preferred metrics: qubit count, QV and CLOPS.
OQC is a UK superconducting qubit startup. In contrast to most other superconducting qubit competitors, OQC has also been using fixed frequency qubits. It’s patented coaxmon qubit architecture has been designed from first principles to alleviate the same routing issues IBM has been addressing with its stacked approach in Eagle. It will be interesting to see how stats compare.
OQC scored a notable milestone by being announced as the first non-North American company to make its processors available via the Amazon Bracket cloud platform. In addition, it has launched the UK’s first commercial QCaaS offering.
Google gathers itself for the next leap
Google plans to spend ‘several billion’ dollars developing a QC by 2029 and is doubling the size of its team. It also now has its own clean room to allow it to fabricate chips with reduced turnaround times.
Google led the field with the implementation of its first milestone: their Sycamore chip demonstrated a calculation beyond the power of any classical computer (quantum supremacy). They are now on schedule towards their second milestone: a logical qubit prototype. Specifically, they want to demonstrate that their preferred error correction approach, the surface code, really can systematically supress errors by moving to encodings of increasing grid size.
In pursuit of this goal, they have implemented upgrades that move the Sycamore architecture from one geared specifically to the artificial problem of ‘random circuit sampling’ towards one optimised for implementing the surface code. This has included upgrading the main 2Q gate they natively implement to one that is much more useful to the task at hand.
iSWAP-like → CZ – sticking with tuneable frequency transmon qubits, the new standard CZ gate is more complicated and somewhat slower than the old gate (14ns → 26ns). It’s a real achievement that it has broadly the same fidelity (99.4% 2Qsim – it is also important to note that Google quote the more stringent test of 2Q fidelity in simultaneous operation, its best individual gate pairs have previously demonstrated 99.9% performance.
Readout & reset – Google has also implemented much faster and higher fidelity readout, moving from 3000ns and 96.2% to 600ns and 98.1%. A new multi-level reset operation is also able to correct leakage errors.
Google’s improvements on readout performance are vital for error correction, because errors on idle data qubits have proved a surprising significant bottleneck . So far, Google has successfully implemented a 1D ‘Repetition Code’ which can be seen as half-way mark on the way to implementing the full 2D surface code .
Lambda – Google emphasise the importance of the parameter that measures how effectively errors are suppressed as we move to larger code distances. Λ greater than 1 shows that systematic error correction is working, but Google argue a Λ of 10 is really the sweet spot for system design. Lower than this the ratio of physical qubits to logical qubits increases unrealistically, higher than this offers relatively diminishing returns. To achieve this point, current surface code protocols are thought to require a 2Q gate fidelity of 99.9%+. Error rates on idle qubits waiting during the measurement and reset cycle also have to be significantly improved. .
Error Bursts – Not all errors are created equal. Error correcting codes have normally been designed to work on the assumption that errors are small and uncorrelated. Leakage errors where the qubit state is destroyed, or errors that are highly correlated across multiple qubits are more difficult to manage. Google points out that mitigating impacts from cosmic rays and latent radioactivity are now a real-world design consideration for further scaling.
Hartmut Neven makes an observation “reaching beyond classical capabilities once and writing a Nature paper is one thing, having it reliably humming over time and providing proper cloud service levels is a whole different thing” . This isn’t a negative, it’s an illustration of how hard practical challenges such as automating calibration can be as devices scale-up. This is a marker for others who may soon face similar problems.
Google roadmap – 100Q (logical qubit prototype), 1000Q (logical qubit), 10kQ (tileable logical modules), 100kQ (engineering scale-up), 1MQ (error-corrected quantum computer) by 2029. Error correction via surface code protocol. Preferred metric: 2Q gate fidelity in simultaneous operation.
QuTech also work with tuneable frequency transmon qubits and have posted striking fidelity results. In a lab test device they have implemented the novel SNZ CZ 2Q gates with fidelity of 99.93%. . This bodes well for the future of the Quantum Inspire open-access cloud. MIT also demonstrated a high fidelity iSWAP 2Q gate with tuneable couplers with fidelity of 99.87% .
The calibration issues that such systems have faced is reflected in work that UKRI is funding in this area. The AutoQT project, led by startup Riverlane, is seeking to leverage AI techniques to keep qubits spinning.
USTC Zuchongzhi makes waves
USTC grabbed headlines in 2021 with Zuchongzhi, a nominally 66Q superconducting qubit device. Zuchongzhi 2.0 was able to execute random circuit sampling with 56Q, Zuchongzhi 2.1 with 60Q, seven more than Sycamore has managed. The net result is a new world record for calculation difficulty.
Zuchongzhi – Named for the Chinese astronomer and mathematician whose accurate calculation of the value of pi around 480 AD was not surpassed for 800 years. This device employs essentially the same architecture as Google Sycamore: tuneable frequency transmon qubits, with tuneable couplers to supress crosstalk, and iSWAP-like gates.
Does this mean USTC is now leading the QC race? Fact Based Insight sees a more nuanced picture. Zuchongzhi’s fidelity stats (99.4% 2Qsim) are now a dead-heat with Sycamore, though its gate speeds are slower than the original Sycamore (24ns vs 12ns). In scientific terms it is a great example of an experiment repeated in another lab to confirm a result. The USTC team now face the same challenges that Google has been finding its way through over the last two years; how to turn a great experiment into a device robust enough for regular operation. The impressive speed with which the readout fidelity was improved between 2.0 and 2.1, with just a 2-month turnaround on a newly fabricated chip, is perhaps an indicator of what we can expect. More challenging will be how to lead on qubit fidelity rather than playing catch-up. Neither USTC nor Google fidelities are yet good enough for what they want to achieve.
Origin Quantum, the Hefei based quantum computing startup, is well placed to benefit as the skills base for superconducting qubit technology grows in China. It already supports cloud access to its 24Q system, with an upgrade to 64Q imminent. Following a successful financing round, with a valuation of ¥7B, Origin also has unicorn status .
Origin Quantum roadmap – 64Q 2021, 144Q 2022, 1024Q 2025.
Superconducting qubit technology in China can expect to benefit from a new state-of-the-art next generation nanofabrication centre being built in Jinan city in Shandong province . This will be a key assistance in an increasingly competitive race.
In Europe, the Quantum Flagship project OpenSuperQ has built initial 5Q and 7Q superconducting qubit devices, also with a tuneable qubit approach. There are prospects of an upgrade to a 20Q flip-chip design before the current effort ends in March 2022 . This is set to be less than the original 100Q ambition of the project. However Fact Based Insight believes we have to be realistic about whether cutting edge projects of this sort can always keep to ambitious timescales. If they always can, our ambition is too low. Equally, the difficulties faced do help to underline the achievement of the USTC team in developing Zuchongzhi.
Rigetti focusses on modularity
Rigetti have had an exciting 2021 in and out of the lab. They made headlines in October by announcing a SPAC based listing in a deal that gives them quantum unicorn status.
Fabricating single chips of increasing scale faces the problem of failed qubits and declining yields. Rigetti has emphasised the development of a modular approach that takes direct aim at this scaling challenge. Small processors fabricated on separate silicon dies are flip-chip bonded onto a carrier chip. The carrier chip incorporates couplers to provide quantum coherent interconnects between the separate chips. Can this approach maintain the required high fidelity? Results in 2021 showed promise. Cross-chip entanglement rates were comparable to Rigetti’s underlying gate speeds (c. 10 MHz); fidelities were promising (99.1% for iSWAP ,98.3% for CZ) .
In the end, many companies may want to use this approach. The cross-over point for optimum fabrication chip-size remains to be established. Significantly improving fidelities is a key outstanding challenge for Rigetti.
D-Wave doubles down
D-Wave laid-out a very significant expansion of their strategy in 2021, announcing plans to introduce a gate-model series of devices. The specifics of their plans are very interesting. They acknowledge that the gate-model approach to quantum computing has more long-term potential for material science and chemistry simulation problems (bringing them into line with the conventional wisdom in the field). However, they are also clear that they are doubling down on quantum annealing as the architecture they believe has the best prospects for optimisation problems, both in the short term, and into the long term (at least versus currently envisaged gate-model architectures).
D-Wave annealing roadmap – 5000Q Advantage 15X connectivity, Advantage performance updates 2022; Advantage 2 new qubit design 2023/24, improved coherence enables improved connectivity 2025.
D-Wave gate-model concept is also distinctive, retaining the flux-qubit design and multi-layer fabrication they have pioneered, including keeping multiplexed control on-chip. Many would have identified these as the compromises not suited to the higher-fidelity specs demanded by a gate-model approach. However, D-Wave has the relevant experts. They didn’t have to define this path if they didn’t believe they could make it work.
D-Wave gate-model roadmap – Phase 1: validate qubits in multilayer stack; phase 2 validate error correction; phase 3 demo logical qubit manipulation; phase 4 Design scalable-task specific components; phase 5 First integrated general purpose processor.
Qilimanjaro is a Spanish startup also pursuing quantum annealing technology. It is participating in the AVaQus project wo build a coherent quantum annealer in Europe.
Flexible entrants push upwards
SEEQC is another startup with long term potential in this segment based on their unique superconducting single flux quantum (SFQ) digital control technology.
Single Flux Quantum – This unique superconducting electronics technology allows SEEQC to integrate fast, low heat & power classical processing, digital multiplexing and low latency qubit control on a single cryogenic chip. Fast classical control logic based on SFQ circuits, in particular for error correction and the efficient production of magic states, could be a key advantage for SEEQC.
Efficient flip-chips – A strength of SEEQC’s approach is that the relative size of superconducting qubits and superconducting control electronics are similar: qubits are located on one wafer/chip, while corresponding SFQ control circuits are on another wafer/chip of the same size. These can be separately fabricated, diced and flip-chip bonded using established fab processes. Such fabrication flexibility combined with wireless coupling promises scaling advantages.
SEEQC have recently fabricated their 3rd generation of multi-chip module test devices including transmon arrays and fluxonium devices in combination with SFQ controllers but are not yet commenting on performance stats.
SEEQC’s integrated qubit control approach has attracted industry co-investment alongside three UKRI grants in 2021. SEEQC’s state-of-the-art in-house commercial foundry capable of producing complex SFQ chips means they are well placed to benefit from political momentum in the US to support investment in chip fabrication technology.
IQM is a European startup with a differentiated offer, focussed on building on-site quantum computers for research institutes and HPC centres. There is a strong niche for this co-design approach, particularly within the EU and potentially independent minded Asian countries such as India. IQM have recently won a high-profile project to build a system in Germany, following up an earlier win in their native Finland .
Bleximo is a US startup also with an emphasis on co-design of application specific quantum computers based on superconducting qubits. Though they have to date kept a low profile, the breadth of their engineering capabilities was on clear display at Q2B. Bleximo naturally benefits from its close connections to the Berkeley research community.
IMPAQT is a Dutch consortium targeting the ‘self-build’ market for research groups.
Future scalability will also at some point probably required coherent connections between separate dilution fridges. ETH Zurich have demonstrated this principle . Development of this technology in Europe has been spearheaded by the Quantum Flagship QMiCS project .
Trapped Ions lead the charge to logical qubits
IonQ shines even though it’s not yet firing on all cylinders
Trapped ions also grabbed many headlines in 2021. IonQ literally rang the bell at the NYSE with its eye-catching $2b SPAC assisted floatation.
IonQ’s academic collaborators also went further, demonstrating fault-tolerant control of a 15Q device to form a single logical qubit (using the Bacon-Shor-13 error correcting code). A full set of 1Q gates were demonstrated, as was the creation of magic states. This is a milestone for the field, though the logical fidelities achieved are only modest (logical SPAM 99.4%, logical 1Q fidelity 99.7%) . Error correction is so often discussed that some might take it for granted. The field should be relieved to see it working in practice.
IonQ also scored an independent benchmarking success. In 2021 QED-C published a suite of performance benchmarks covering a range of common underlying quantum algorithms. Results included IonQ’s cloud-accessible 11Q system and also IonQ’s next generation hardware. The latest IonQ hardware is a big step forward from the previous generation and overall it outperforms the other devices on test from IBM, Honeywell and Rigetti. The informative visualisations developed by QED-C allow the benefits of the enhanced qubit connectivity offered by the trapped ion systems to be clearly demonstrated.
QED-C Benchmark Suite – This takes a wide selection of commonly discussed quantum algorithms and presents their performance at different qubit scales and circuit depths. The visualisations produced are beautiful and very informative. They support an intuitive link to QV and provide natural insights beyond a simple number.
However, the performance of IonQ’s new hardware is also in some ways disappointing. It was announced last year as a 32Q device, but seems to have run here in a 21Q configuration. The measured QV of the system has not yet been published, but Fact Based Insight’s estimate is that at the time of the benchmark it appears to have been operating in the 1024QV range, still significantly short of the 4M QV we want to see. IonQ seems to be facing delays in tuning up the 2Q gate fidelity to the full potential of the system.
IonQ roadmap – 22AQ 2021, 29AQ 2023, 64AQ 2025, 256AQ 2026, 384AQ 2027, 1024AQ 2028. Error correction – 16:1 2025, 32:1 2027.
Algorithmic qubits (AQ) – IonQ has defined this measure to indicate the number of ‘effectively perfect’ qubits available for calculation (Note that available logical gate depth will still be limited). In the absence of error correction encoding AQ=log2(QV).
A key part of IonQ’s strategy is to take a flexible approach to error correction. Rather than wait to implement a high-performance high-overhead surface code, it emphasises medium term plans to make pragmatic use of smaller codes. This promises a useful increase in fidelity (perhaps to 99.99%) with only a modest overhead (16:1).
The superconducting gauntlet – IBM’s 99.91 ± 0.014%, QuTech’s 99.93 ± 0.24% and MIT’s 99.87 ± 0.23% 2Q gate fidelity results with superconducting qubits is a significant challenge to the trapped ion community. A key part of the trapped ion story is that they offer significantly higher 2Q gate fidelities. However that record is only 99.92 ± 0.04% for laser driven gates , and 99.91 ± 0.09% for near-field MW driven gates ; and those levels of performance have only been demonstrated in simple 2Q lab experiments. No trapped ion player has demonstrated anything close to this in a true multi-qubit device. The IBM result is effectively a dead heat on fidelity and is in a production fabricated system. Superconducting qubit gate speeds are also significantly faster.
Ion trap systems do have further cards to play. Enhanced qubit-to-qubit connectivity may be very relevant for potential NISQ applications (superconducting qubits are typically restricted to nearest neighbour interactions). This may also translate into important long-term benefits if it allows access to innovative error correcting code implementations with reduced overheads. In the variability of superconducting qubit gate fidelities, we can also see direct evidence of another issue often emphasised: fabricated qubits suffer from an inevitable variability that ion or atom based qubits don’t.
At Q2B 2021 IonQ announced new details of its technical roadmap. Its next generation processors will use barium ion qubits, rather than the ytterbium ions used in its current devices . Fact Based Insight doesn’t expect an immediate leap in performance, but overall, this speaks very well of the health of IonQ’s technical vision. A subtle challenge of having a $800M investment war chest is whether you have sufficient good ideas for how to put the money to work. IonQ’s commercial and scientific leadership seem up for this opportunity.
The Goldilocks qubit – 133Ba does not occur naturally. Instead, this man-made isotope has been synthesised specifically because of its promise as a trapped ion qubit. It has a nuclear spin of ½ and so can be used as a base for long lived hyperfine qubits. Crucially, the electronic structure of the ion also has metastable states with transitions at convenient optical wavelengths. These are much easier to work with in optical fibre or integrated photonic devices. Such transitions can be used for more convenient manipulation of the hyperfine qubits, or as a base for optical qubits to be formed in the same system. This also promises to simplify the task of creating photonic interconnects between traps . However, experimental work with barium ions is still at a relatively early stage compared to other ion species . We don’t know if IonQ will immediately switch to this particular isotope of barium, but Fact Based Insight believes this may be where it will move to in the end.
IonQ place significant emphasis on the modularity of their approach. Chief Scientist Chris Monroe points out that even before moving to multiple devices, ion-chains can be moved and combined within a single extended trap to create modules of a few hundred qubits. Integrated optics are a key requirement for scaling, but he points out that this is an engineering challenge, not a physics one.
AQT demonstrate a true rack based system
AQT are a startup based on trapped ion optical qubits (ones based on optical transitions rather than the hyperfine transitions used by IonQ and Honeywell). These offer somewhat lower coherence times, but promises easier future development of an integrated optical system. AQT have leveraged their involvement in the EU QT Flagship AQTION project to deliver a first of its kind 24Q fully rack-mounted demonstrator system – the smallest QC available in the world today .
Work presented at EQTC 2021 shows the AQTION and University of Innsbruck team demonstrating a full universal set of fault tolerant gates (1Q gates, 2Q gates and magic state injection). This world first uses a 16Q AQT Pine Trap device to implement 2 logical qubits (each using a 7Q color code, also known as the Steane code). Thomas Monz led the Innsbruck team and is also co-founder of ATQ. If this result withstands peer-review, it will be a significant win for the European programme .
The 2Q fidelity demonstrated by Pine Trap is 99.36% . Improving this further will be a key challenge.
Light-Shift 2Q gate – Work at Georgia Tech points to the potential advantages of a new mechanism for laser driven 2Q gates. This can in principle be applied to hyperfine or optical qubits. Recent work argues that this approach is capable of supporting gates of up to 99.99% fidelity with optical trapped ion qubits.
Honeywell becomes Quantinuum and continues to deliver
Honeywell made a big corporate move in 2021, spinning off its quantum computing division to merge with Cambridge Quantum, a world leader in quantum software, and forming a new group called Quantinuum. Direct access to leading soft-side knowhow will be an important differentiator in getting the most from early devices. Continuity on hardware development is ensured by the strong capitalisation of the deal, continuing access to Honeywell fabrication facilities and the continuation of Tony Uttley as President and COO of the new entity.
Honeywell continue to demonstrate their ability to move systematically forward with their roadmap. In 2021 they again set records by being first to reach 1024QV and are now targeting ‘order of magnitude’ increases in QV each year.
Preliminary results from Honeywell also show successful multi-round error correction within a 10Q device (using the [[7,1,3]] color code). Honeywell has used its lead on mid-circuit measurement to allow it to efficiently repeat multiple rounds of error correction. Logical fidelities are up on physical fidelities (logical SPAM 99.83(2)% vs physical SPAM 99.76(8)% ). However, the detailed error analysis in Honeywell’s work also draws attention to the need to better understand and improve underlying physical error rates so that error correction doesn’t have to run so hard just to stand still . Many early players will soon also face this challenge.
Honeywell roadmap – H1 (linear trap), H2 (racetrack layout), H3 (grid layout), H4 (integrated optics), H5 (large scale via tiling); by 2030.
Honeywell’s approach to ion traps is also very different to that of IonQ or AQT. Rather than manipulating a linear chain of ions, Honeywell shuffles ions in a QCCD grid. Other developments in 2021 point to interesting new possibilities for this type of architecture.
Innovation on trapped ion interconnects
Universal Quantum are a trapped ion startup also using a QCCD approach. A traditional question for proposed large scale trapped ion architectures is what to do when you’ve built the largest feasible single trap. This might be sufficient for a small NISQ application, but it certainly wouldn’t be enough for the million-qubit devices expected to be required for FTQC. Universal propose a novel solution to this problem, the direct transport of ions between modules.
Photonic Interconnects – The standard proposal for coherently linking modules together is to use photonic interconnects, and indeed proponents point to the flexibility of this solution as a strength for trapped ion architectures overall. However, the speeds demonstrated so far with such links are modest (entanglement rates of 182 per second with fidelity 94% – if you are old enough, you may remember trying to access the Internet via a 320 baud modem on a bad line).
Direct transport – At ECTI 2021, a team from University of Sussex and Universal announced a striking proof-of-concept demonstration of ‘all electric’ ion transport across micro-aligned quantum module boundaries. This offers a large improvement in rate and remarkable transfer fidelity (2400 per second and 99.999995% – if you are old enough you will remember that 2400 baud really was better than 320).
Direct transport isn’t as flexible as a photonic link, but it offers a tool that could combine well with the needs of error correction. Others point to their expectation that the speeds of photonic interconnects will increase substantially. Entangled Networks are a Canadian startup focussed on this technology. Aharon Brodutch (CEO) envisages rates of 1000 per second being within reach.
Innovation on trapped ion gates
Universal Quantum is also one of the trapped ion players who continues to question the scalability of driving gates directly via lasers, as used in the IonQ, AQT and Honeywell systems. In Universal’s case they are pursuing gates driven by global microwave fields. This promises unique scaling benefits, but only if gates of sufficient fidelity can be achieved and sustained for large numbers of ions.
Universal has been successful again in 2021 in winning additional grant funding from UKRI. One project will support the development of an error corrected quantum processor for commercial applications, a second supports microchip development with major semiconductor manufacturers.
Oxford Ionics is a trapped ion startup focussed on a completely different near-field microwave approach to driving gates. Spinning-out from the UK’s QCS hub, it has more recently benefitted from a series of UKRI grants, including three more in 2021. It has not launched a device yet, but its founders have built a series of record-breaking lab systems. In 2021 it has been hiring top talent.
Beyond 99.99% – Oxford Ionics co-founder Tom Harty strongly makes the case that trapped ions have the potential to go well beyond 99.99% fidelity and restore clear blue water between them and superconducting qubit systems. At its core, this technology platform allows us to combine the advantages of two of our most powerful tools: the repeatability of modern semiconductor fabrication techniques; and the same natural qubit systems at the heart of atomic clocks, literally the most precise devices humankind has ever built.
Oxford Ionics only use lasers for controlling classical aspects of the system (cooling and readout). Harty points to the advantages of driving delicate quantum gates with microwaves, arguing that phase sensitivity should be easier to deal with at these MHz frequencies, rather than the THz seen in typical laser systems. Also, microwaves could provide a way past the photon scattering limit that threatens to make it difficult for laser driven gates to get beyond 99.99% .
We can expect to see Oxford Ionics announce separate NISQ and FTQC roadmaps. If sufficiently high 2Q fidelity can be achieved why wait for error correction? 99.99% 2Q fidelity might allow a 100AQ device. That would be very interesting.
Photonics reveals its alternative vision
PsiQuantum lets in some light
PsiQ raised another $450M in 2021 for total funding of $665M and a valuation of $3B. PsiQ also emerged from stealth with specific details of their approach.
PsiQ have now been working with GlobalFoundries, one of the world’s leading semiconductor manufacturers, for some time and this partnership has begun to demonstrate impressive results. A crucial step has been the incorporation of tooling for PsiQ’s single-photon source and single-photon detection technology (Niobium Nitride SNSPD), together with in-line testing (4K cryostat based) into the process lines at GF’s state-of-the-art 300mm production facilities in upstate New York and Dresden, Germany.
Tier-1 fabs – A crucial benefit is the improved nano-feature accuracy that top tier facilities can produce. Line-edge roughness in the waveguides leads to photon loss and impairs detector efficiency. PsiQ reports that moving from boutique to state-of-the-art facilities has led to a five-times improvement in patterning quality, leading to huge gains in device performance (for example, single photon detector efficiencies moving from 97%@2K to 99.7%@2K. Importantly thousands of detectors can be fabricated on each wafer alongside the sources, waveguides and other optical components needed for the system.
PsiQ has pulled-off a delicate balancing act. They have demonstrated that key components of the integrated device can be improved by leveraging the capabilities of top-fabs. At the same time they have avoided exotic materials and ultra-cutting edge manufacturing processes that would have precluded their access to such fabs, who’s main business is not R&D but the bulk production of conventional chips.
Jeremy O’Brien (CEO of PsiQ) summarises “Rather than trying to scale a quantum process, we’ve taken a scalable process and have made it quantum”.
Fact Based Insight sees PsiQ’s theoretical breakthroughs as even more innovative and potentially far reaching. PsiQ use single photons to form qubits. They have now introduced what they call fusion based quantum computing as a significant simplification of how fault-tolerant calculations can be supported in this architecture .
Dual-rail photon qubits based on photon path – The |0> and |1> qubit states are encoded in the path taken by the photon: is it in the upper or lower waveguide or both.
In this scheme, the conventional 2Q gates of circuit-model quantum computing are not practical. Instead an equivalent scheme called measurement-based quantum computing has often been proposed.
MBQC uses a large entangled ‘cluster state’. The required quantum algorithm is implemented via a series of measurements on the cluster state. A key challenge for this approach has always been that sufficiently large cluster states are difficult to produce.
FBQC takes as its input small standard entangled states. These resource states are then combined (fused) on the fly using standard photonic techniques (Bell measurements). This effectively recreates the power of a fault-tolerant cluster state, but crucially the number of optical components that any single photon has to successfully navigate is kept low and constant.
Fusion Based Quantum Computing challenges some of our normal intuition about what matters in quantum computing. In conventional circuit-based approaches we typically focus on the notion that the coherence of the qubit system has to be protected through an increasing number of gates. This allows the quantum correlations that quantum algorithms exploit to build up in the final qubit states. In FBQC each individual qubit/photon is measured and destroyed shortly after its creation. The quantum correlations build up in the classical data associated with the control of this process.
An elegant feature of the FBQC approach is that multiple forms of potential error (photon loss, fusion failure etc) are naturally resolved as part of a unified error correction process. This is closely related to the topological ideas at the heart of approaches such as the surface code. In some sense the multiple rounds of fusion in FBQC are analogous to surface code error correction cycles. In this sense both take a 2D structure and add time steps to create a ‘3D’ fault-tolerant channel along which the computation can proceed. However, the flexibility with which the photonic architectures can be fabricated promises to combine well with the theoretical flexibility of the FBQC approach. While the ‘example’ architectures PsiQ have published use a 6Q entangled ring as their basic resource state, they hint that this is not optimal and enhanced performance may be possible using different unit cells shapes. This is exactly the area of error correction theory for which members of the PsiQ team are well known.
What are the potential weaknesses of the PsiQ approach? We haven’t yet seen devices in operation. Just as with other technologies we can expect real experience to bring a keener sense of specific challenges. Just as with any QC platform, PsiQ’s qubits are ultimately analogue objects and so a fabrication tolerance or calibration challenge must be faced somewhere. For path encoded photonic qubits this could exhibit itself as a sensitivity to photon phase. Crosstalk is a challenge for other platforms but is normally considered a likely strength for photonics. However, no one else has ever worked with single photons at the scale and component density that PsiQ now envisage. PsiQ also have not said much on the technology they are using to generate the initial entangled resource states on which their approach depends. Each resource state generator is likely to require significant switching and multiplexing complexity.
Classical control – A strength of PsiQ’s approach is that the relative scale of photonics fabrication and control electronic fabrication works out nicely: about 1 million qubits fit on a 300mm wafer, while about 1 trillion transistors fit on a (differently optimised) wafer of the same size. Such chips can be separately fabricated, diced and then flip-chip bonded using established fab processes. This results in a package with a very convenient control electronics/qubit ratio.
A further remarkable advantage of the FBQC architecture is that much of the basic control processing doesn’t have to happen within the physical photonic clock cycle; instead, the conventional calculation can ‘catch-up’ after the end of multiple fusion cycles. This feature extends even to the creation of magic states, a key resource needed here (and in most FTQC schemes) to achieve quantum advantage.
In Fact Based Insight’s view, classical control logic also remains a watch point. PsiQ’s detectors, and so their whole photonics chips, still need to be operated at cryostat temperatures. 2K is 100 times hotter than 20mK and so a much easier temperature to work at than those faced by superconducting qubits. However, a significant part of the control logic is also going to have to run at these low temperatures. Such cryo-CMOS technology is still itself an emerging field. True, many other quantum hardware players face this challenge, but it’s PsiQ that is claiming it will have built its system by mid-decade, so its PsiQ that needs to be on point on this issue.
Classical control logic, in particular to produce magic states, could be a key bottleneck for PsiQ. We should be cautious in assuming that fast photon flight times automatically mean fast logical cycle times. The speed of the classical processing will most likely dominate.
PsiQ roadmap – 1MQ providing 100-300 logical qubits, within 5-8 years. Error correction via FBQC.
For the longer term, PsiQ would like to move away from SNSPD technology so that their devices could run at ambient temperature . They are in a great position to attack this challenge from both ends – new hardware technologies that don’t require cryogenic cooling (such as are being used in the latest generation of QKD solutions) and improved error correction protocols, allowing a higher degree of error to be tolerated.
USTC Jiuzhang 2.0 moves onwards
USTC turned heads in 2020 with the initial results from Jiuzhang. In 2021 these results have been extended to further increase the complexity of the calculation and to introduce a degree of programmability .
Jiuzhang – Named for the Nine Chapters on the Mathematical Art, a famous Chinese text composed between the 10th & 2nd century BC. The initial device detected up to 76 output photons across 100 output modes. This has now been upgraded to 113 output photons across 144 output modes.
USTC vs Google – the debate over the correct interpretation of the Jiuzhang results has provided a great illustration of the benefits to all parties of transparent competition science-style. Google responded to the USTC results demonstrating that in the presence of experimental noise, ‘noisy’ boson sampling could be classically simulated much more efficiently than previously realised. In particular they could recreate the specific metric originally proposed by Scott Aaronson as a test of quantum supremacy more efficiently than Jiuzhang. USTC countered that high-order quantum correlations (that can’t be simulated by the classical approach) were still there in the Jiuzhang data. The result is that all parties, even Scott Aaronson, now understand boson sampling better than before.
This is of more than just scientific interest. It gives us an intuition about how difficult it will be to demonstrate practically useful quantum advantage using boson sampling on a noisy NISQ era device.
The real limitation of Jiuzhang 2.0 remains that it is based on discrete components in a conventional optical table setup. This approach is not scalable. The obvious path is to integrate the components in a way similar to that used by Xanadu or PsiQ. Mainland China has for the last decade been investing strongly in building-up its semiconductor fab industry, moving from a minor participant to a top-5 player . Hefei, where USTC is based, is a notable centre for 300mm wafers in China, with one Nexchip fab in operation and three others reportedly planned .
Xanadu details its own path
Xanadu raised a further $100m in 2021 to continue its own journey towards photonic quantum computing.
Rather than single photons, Xanadu uses squeezed light to form its qubits. This allows sources and other photonics components to be combined on chip using standard SiN photonic fabrication technology. Only the detectors require cryogenic cooling and are currently kept off-chip for that reason.
In complete contrast to PsiQ they have focussed on making early devices accessible to customers. Since their first 4Q system was launched in 2019, they have been roughly doubling system size every 6 months. In 2021 they have a 12 mode system available online, a 24 mode system integrated and a 40 mode system in testing (in Xanadu’s scheme for quantum computing, modes play a similar role to qubits).
In 2021 Xanadu published further details of its own unique roadmap to FTQC via GKP qubits . A key advantage is that this scheme has inbuilt resilience to photon loss even before further error correction. Fact Based Insight expects FTQC to increasingly be Xanadu’s focus. Challenges include multiplexing at key bottlenecks, and support (probably by FPGA) for the classical processing needed to run error correction. Opportunities include taking advantage of enhanced 3D connectivity.
Xanadu is based in Canada, but has strong links to the SiN photonics ecosystem in Europe. It has recently partnered with imec for chip production and VTT for superconducting photon detectors.
Xanadu roadmap – Three processor series in parallel 2021-23: X-Series X24, X40, X80; XD-Series (100% connectivity) XD4, XD8, XD12, XD40, XD80; TD-Series (time domain multiplexing) TD2, TD3; 5kQ module (by 2024), FQTC data centre (2025+); Error correction via GKP qubits.
QuiX get on with selling
QuiX a Dutch startup, already make photonic processors with 4,8,12 modes, including the required peripheral control electronics. These leverage the SiN photonics platform to provide leading low-loss performance . QuiX have growing commercial traction and are one of only a small number of companies to have actually sold quantum computing hardware. Testing on an expected 20×20 mode device is rumoured to be complete but no launch announcements has yet been made.
QuiX roadmap – 20×20 mode (2021), 50×50 mode (2022), computational advantage 2023
Orca computing are another interesting photonics startup in the UK. Their approach emphasises their proprietary memory technology which gives them much more flexibility in how they manipulate photonic states. They are combining this with standard optical components to offer a unique platform. The flexibility of this approach is well matched to the UKRI grant funding they have won looking at applications in the quantum data centre of the future.
Quandela, a French startup known for its novel quantum dot single photon source technology, has raised €15M to build a photonic quantum computer. An initial device is expected in 2022.
Duality Quantum Photonics continue to keep a low profile. Given the track records of two of the founders, Anthony Laing and Alberto Politi, it’s a no brainer that their plans are going to involve developing integrated photonic quantum tech (Politi’s PhD thesis ‘Integrated Quantum Photonics’ initiated the field back in 2008; his then co-supervisor was Jeremy O’Brien).
Neutral Atoms lead on analogue simulation
Systems based on neutral atoms also continued to generate excitement in 2021. A growing number of companies are pursuing this technology, including ColdQuanta, QuEra, Pasqal, Atom Computing and M Squared.
QuEra took a large step in 2021 with the demonstration of a 256-atom analogue quantum simulator. This 2D array of cold atoms doesn’t yet implement a full qubit gate-set, but it can already be programmed to recreate the dynamics of analogous quantum systems. The results are already of scientific interest, and potentially have wider applications .
Pasqal is a French startup also pursuing this path. Its founders have demonstrated a 200-atom analogue quantum simulator . Work at Pasqal is emphasising what can be achieved with this modality (both at pulse control and algorithmic levels), as well as progressing towards a gate-model device.
Atom Computing, another startup setting out on this path, has unveiled a 100-atom simulator and made high profile hires in 2021.
ColdQuanta has been an early leader in neutral atom technology. Its work on applying this technology as a base for quantum computing is supported by the DARPA ONISQ programme. It looks like it will just miss its goal of getting a 100Q device on the cloud in 2021, but this is now expected to follow early in 2022.
QuEra Roadmap – 256 atom simulator (2021), 64Q (2022), 1024Q (2024); Error correction using Steane-7
Pasqal roadmap – 200 atom simulator (2021), 1000Q (2023)
ColdQuanta roadmap – H1 100Q 4X connectivity (2021, now early 2022), H2 200Q 8X, H3 300Q 20X, H4 1000Q 50X (2024), H5 2000Q+ 60X+
Neutral atoms systems have stolen a lead as analogue quantum simulators. The nature of their traps also give a clear route to scaling to about 1000Q. Further than this, ColdQuanta envisage that a device with 100,000Q in a 1mm2 footprint is possible, though concede that managing the complexity of the required optics systems will be a big challenge.
The potential of such systems to operate as universal gate-model quantum computers received a boost in 2021 with the demonstration of 99.1% 2Q gate fidelity at Caltech . This uses highly excited ‘Rydberg states’ of the atoms. However, the best work of which Fact Based Insight is aware does not yet show the path to 99.9%+ . Readout, and especially the reset cycle time, may also be a challenge for this technology. In general, the technical path to FTQC is not so well established here.
It’s important to realise that neutral atom technology has a strong presence in other areas of the quantum technology ecosystem, in particular in quantum sensing. This provides diversity and supporting opportunities for players in this area.
M Squared, known for its high-performance lasers, has already established a strong position enabling this technology for quantum sensing. In 2021 it has also started to build out its own quantum computing system.
Silicon Spin qubits make a triple jump
Silicon spin qubits have long had an intrinsic appeal. Isotopically purified 28Si offers a ‘spin vacuum’ in which to host the delicate spin qubits. It also holds out the promise of leveraging the fabrication techniques developed for the semiconductor industry. However, the fidelities achieved with delicate spin qubits had lagged other platforms.
In a big step forward for silicon spin qubits three teams have now announced results beyond 99% 2Q fidelity.
QuTech have demonstrated 99.5% 2Q gate fidelity in a 2Q test device (a gate-defined double quantum dot in an isotopically enriched 28Si/SiGe heterostructure) . The ability to connect two such quantum dots has also been demonstrated .
RIKEN have demonstrated 99.5% 2Q gate fidelity in a 2Q test device (a triple quantum dot again in an isotopically enriched 28Si/SiGe heterostructure) .
UNSW have demonstrated 99.37% 2Q gate fidelity in a 3Q test device (a pair of ion-implanted 31P nuclei on a silicon device allow an electron spin and two nuclear spins to be coupled. A real proof of concept for the scalability of donor qubits) .
The technical and geographic diversity of these results is a striking indication of the potential of this qubit platform. The profusion of techniques being explored by academic groups are not always mutually compatible, but together they represent a strength in possibilities. Several other interesting technologies were demonstrated in 2021.
Spin qubit gates driven by global microwave fields – UNSW demonstrated spin qubit gates using the same technique being pursued by some trapped ion players due to its potential scaling advantages . This technique traditionally makes use of ‘dressed states’ to supress sensitivity to magnetic field noise. In pursuing the technique for silicon, the team at UNSW have introduced a new enhanced ‘SMART’ protocol that promises higher order noise suppression .
Germanium – A different approach is to define qubits based on holes (rather than electrons) in a germanium layer on a silicon wafer. This is emerging rapidly as a promising platform: the first germanium qubit was realized in 2019, two-qubits in 2020, and a four-qubit system has been demonstrated in 2021 . Progress in germanium is a proof of concept that semiconductor technology can be used to scale qubits in two-dimensional arrays.
Photonic Inc – Continue to work to bring photonic interconnects to the silicon spin platform.
Intel reminded everyone in 2021 of their long-term potential strength in spin qubits. In joint work with QuTech they demonstrated the fabrication of quantum dot arrays in a state-of-the-art 300mm fab. Initial proof of concept devices have already achieved 1Q gate fidelities and coherence lifetimes comparable with previous test devices . The scaling potential is tantalising. As Intel says, a quantum dot is really just like the extreme one-electron case of a transistor. On the other hand, the potential crosstalk issues in larger devices have not yet been explored.
Lithography – Intel identifies adherence to the strict fab design rules as key to being able to achieve the extremely high yields required. In particular it points to the use of all optical lithography (inc. chemical mechanical polishing) rather than the electron beam lithography often used to date for quantum dot fabrication. A single 300mm wafer contains 10,000 quantum dots arrays, with up to 55 gates per fin.
Cryo-CMOS – Intel and QuTech have also successfully benchmarked the latest Horse Ridge control chip capable of operating at 3K to drive microwave control of qubits.
UNSW spinout, SQC is perhaps the best known silicon qubit startup. However, it uses a very different flavour of this technology: phosphorus atom qubits in silicon. These devices are fabricated using an ultra-cutting-edge technique offering atomic precision way beyond conventional CMOS techniques. We will have to wait some time yet to see if their vision can indeed leapfrog the field.
SQC roadmap – 10Q prototype by 2023, 100Q before 2030, useful FTQC by the mid-2030s.
In Europe, the launch of the QLSI project is a significant boost for silicon spin proponents, drawing in expertise from CEA-Leti, CNRS, imec and others. Key targets include making an 8Q processor available via QuTechs Quantum Inspire open-access platform and demonstrating a 16Q chip by 2023.
Quantum Motion, a UK silicon spin startup can expect to benefit from its involvement in QLSI. It has also won funding from UKRI to develop cryo-CMOS solutions for control and readout.
Carbon holds its own unique potential
NV centres in diamond are another quantum information platform. 12C has no nuclear spin so forms a magnetically neutral host for qubit systems. Diamond also has great thermal properties that mean devices can retain performance even at room temperature (though fidelity improves at lower temperatures). The fidelity of defect-centre to defect-centre gates remains a challenge. However multiple 13C nuclear spins associated with a single defect can deliver high fidelity performance. The favourable optical properties of this platform means that it is often considered a candidate for small processor nodes on the Quantum Internet.
QuTech – In a striking 2021 demonstration, entanglement was distributed between three NV Diamond nodes (cooled to a temperature of 4K). Each contains one qubit used for communication. The middle node also uses a 13C nuclear spin as a memory qubit.
Quantum Brilliance are an Australian startup building room temperature quantum computers based on NV diamonds. Their Gen1 desktop system is already commercially available.
The remarkable properties of NV diamonds are also allowing them to make rapid strides in the field of quantum sensing. We can expect to hear a lot more about this technology.
Archer, an Australian materials technology company, has also been preparing its entry into the quantum technology market. Its carbon nanosphere technology offers a new substrate option for spin qubits, but is at an early stage of development. Qubit lifetimes at room temperature (175ns @ 300K ) are less than for NV Diamond, but this platform promises to bring new manufacturing flexibility. Qubit logic operations have not yet been demonstrated. Archer have secured a series of patents in Asia, Europe and US.
C12 Quantum Electronics are building spin qubits in isotopically purified carbon nanotubes. Again manufacturing flexibility is a key advantage. Operation is expected to target the dilution fridge regieme for maximum performance.
EeroQ are developing another novel form of spin qubit based on electrons trapped over liquid helium. Development is at an early stage.
Cat qubits start to take shape
The AWS team has assembled very significant bench strength in quantum error correction. One result has been increasingly detailed proposals for FTQC architectures with significantly lower overheads, including in particular via cat qubits . Other work is pointing to more efficient techniques for practical surface code implementations .
Alice & Bob are a French startup that have already set out on the path to realise cat qubits.
The right stuff
The race continues to deliver devices of sufficient power to unlock quantum computing’s remarkable potential. While clear leaders have emerged, each face very different scaling challenges. It’s still not clear which major qubit technology platform will succeed either in the NISQ or FTQC era. The winner for FQQC applications on the Quantum Internet may be different again, and this may become even more significant if progress towards commercial NISQ applications stalls. The diversity of flavours of each technology still being developed leaves plenty of room for startups to shine.
Fact Based Insight believes that delivering on underlying qubit performance is now of prime importance. In circuit-model terms, 2Q gate fidelities of 99.9% are set to become a minimum bar and 99.99%+ an even more uncomfortable target for those seeking broad quantum advantage in the NISQ era. Results for isolated gates won’t be enough (even when validated via randomised benchmarking). We need to see results in large devices with gates in simultaneous operation (validated via advanced techniques such as cycle benchmarking ).
For those seeking to tread the path to FTQC, you need to be able to explain very clearly how your approach to error correction works not just in principle, but also for the errors present in your real device. Quantum hardware needs to scale, but so too does the required classical control hardware.
The conventional circuit-model community also needs to understand it is not the only game in town. Heuristic approaches such as quantum annealing have significant momentum. Analogue quantum simulation is still virtually unexplored outside of specialist applications in physics. Simple ‘few-qubit’ devices can have novel applications on the Quantum Internet. Perhaps most disruptively, new error correction inspired approaches, such as FBQC, may be pointers to what can be achieved if we think beyond the conventional paradigm.
The quantum future is bright, but it remains full of uncertainty.
To watch in 2022
- IBM – will the 127Q Eagle join the ‘beyond classical’ club?
- Google – will we see the surface code ripple outwards in anger for the first time?
- Rigetti – what specs will we see from its new UK-based computer?
- Zuchongzhi+ – will China capture a lead in error correction and fault tolerance?
- OriginQ – will a large Chinese superconducting qubit device rattle the cloud?
- OQC – what performance stats will we see from its new 8Q cloud processor?
- IonQ – how high a QV can its 2nd gen Ytterbium processors reach? Watch out also for early performance results on 3rd gen Barium processors.
- AQT – what direction next for ‘the smallest QC in the world’?
- Honeywell – watch out for demonstrations of error correction across a universal gate set.
- Universal – with its high-magnetic field chips working; what fidelity can they hit?
- Oxford Ionics – will we see a first near-field microwave trapped ion device enter the race?
- PsiQ – will we hear more about the entanglement sources they have promised to sell?
- QuiX – watch out for the launch of their 20×20 mode device. What performance stats will we see?
- Xanadu – their 24×12 mode device has just hit the cloud; time for a mode-metrics battle?
- Jiuzhang+ – watch-out for signs of which flavour photonics platform the USTC experts prefer.
- SEEQC – when will we get a first glimpse of its SFQ technology in action?
- Neutral atoms – will atom simulators or gate-model devices rule the neutral atom cloud?
- Silicon spin – will we see progress towards true multi-qubit arrays?
- Diamond – will we see progress on unique FQQC applications?
- Cat qubits – will Alice & Bob start to move cats up the agenda? Will we see stats for AWS’ first hardware? Will we see cats from QCI’s stellar team?
- Topological qubits – after the setbacks of 2020/21; can the story move forward again?
- Innovation – watch for qubit control logic demonstrations from emerging players such as EeroQ, Archer’s or C12.
- EU QT Flagship – what winners will emerge from the next round of Horizon Europe projects?
- UK NQCC – the new national quantum computing centre will finally open. What it chooses to support will be a barometer of wider progress.
- US NQI – what new advances will emerge from the NSF and DOE National lab centres?
- CEA-Leti – can this major skills hub start to accelerate the silicon spin qubit journey?
- QuTech – what new processors will we see on Quantum Inspire?
- China – will we see ‘catch-up’ demonstrations also on trapped ion or neutral atom systems?
- Metrics – don’t expect the trapped ion folks to be as quick to adopt CLOPS as they were to embrace QV; but can users at least insist on the new QED-C Benchmark suite?
- Fidelity – who has what it takes to get convincingly across the 99.9% 2Q gate fidelity line?