Is it possible to get quantum advantage without actually using a quantum device? One Swiss start-up, Terra Quantum, says it’s not only possible but that doing so is a core part of its business model, which uses clever software and workflow to simulate qubits on classical HPC resources. There’s a good deal more to Terra Quantum’s strategy than just using simulated qubits in applications – it is, for example doing research on superconducting hardware qubit technology, and its companion company, QMware, looks a bit like AWS Braket delivering access to a home-grown simulator and to third-party QPUs.

Markus Pflitsch is a founder and the CEO of both Terra Quantum and QMware and at first glance the arrangement can seem confusing. QMware is jointly owned by Terra Quantum and classical HPC resource supplier, Novarion Systems.

Founded in 2019, “Terra Quantum is a quantum as a service company structured in three business units: algorithm, compute and security. We are a full-stack company doing software and hardware. On the hardware side, we work on a novel QPU technology based on superconducting qubits. It’s a novel approach [based on the] expertise of Valerii Vinokur [who] is our CTO based in Chicago [and long-time researcher] with the Argonne National Lab,” explained Pflitsch. “But from the 30,000-foot level, we’re a hardware-agnostic software company. We are very strong in developing algorithms for optimization, simulation, and machine learning topics. We run our software and try it out on any available quantum device we can get access to [whether] it is a native QPU or a simulator.”

The shortcomings of today’s noisy intermediate scale quantum (NISQ) systems, says Pflitsch, prompted creation of QMware in 2020. “Today’s quantum hardware isn’t really capable of solving meaningful problems, right? NISQ devices have too many errors. So how should we overcome kind of this NISQ bottleneck to deploy our quantum software? The solution was [use of] simulated qubits on HPC. We formed a joint venture actually, QMware, to do that,” he said. “[It’s] a cloud offering where we offer our simulated qubits, but also access to other quantum devices with a web integration or also in the physical integration.”

Despite its youth, Pflitsch emphasized, “Terra Quantum is one of the largest global quantum tech players. We have more than 150 people [including] more than 130 quantum physicists and more than 750 years of combined relevant quantum tech experience. We raised close to $100 million so far. [Terra] is very European-dominated in terms of its investors, we have two of the richest German families on the cap table. [We are] deeply European rooted, but of course, North America is an important market [too].”

Keeping the two companies separate, said Pflitsch, is intended to make QMware an attractive platform for other quantum companies – he mentioned Multiverse as an example – to deliver their products and services from while Terra concentrates on its IP (software and hardware).

Much of the Terra-plus-QMware pitch is familiar. The commercial quantum landscape is expanding rapidly with many new entrants, some betting on specialization and others, such as Terra Quantum, tackling “full-stack” approaches. Amid the many jostling quantum puzzle pieces, there is an emerging consensus that the end ‘quantum product’ is likely to be a hybrid system in which portions of application are executed on classical systems and portion are accelerated on quantum devices.

An important intermediate step in quantum’s development has been the development of quantum simulators (hardware and software) that mimic quantum device capabilities (superposition and entanglement). For example, IBM, Atos, and AWS, all have qubit simulator offerings. While these simulators are generally thought of as development tools for getting ready to use genuine quantum hardware, Terra/QMware says its quantum simulator is not only among the best available but also being used today by clients to deliver quantum advantage.

To buttress its position, the company recently released a study – Benchmarking simulated and physical quantum processing units using quantum and hybrid algorithms – which compares the performance of QMware and AWS simulators as well as performance against QPUs from IonQ, Oxford Quantum Circuits, Rigetti, and IBM.

Here’s the abstract:

“Powerful hardware services and software libraries are vital tools for quickly and affordably designing, testing, and executing quantum algorithms. A robust large-scale study of how the performance of these platforms scales with the number of qubits is key to providing quantum solutions to challenging industry problems. Such an evaluation is difficult owing to the availability and price of physical quantum processing units. This work benchmarks the runtime and accuracy for a representative sample of specialized high-performance simulated and physical quantum processing units.

“Results show the QMware cloud computing service can reduce the runtime for executing a quantum circuit by up to 78% compared to the next fastest option for algorithms with fewer than 27 qubits. The AWS SV1 simulator offers a runtime advantage for larger circuits, up to the maximum 34 qubits available with SV1. Beyond this limit, QMware provides the ability to execute circuits as large as 40 qubits. Physical quantum devices, such as Rigetti’s Aspen-M2, can provide an exponential runtime advantage for circuits with more than 30. However, the high financial cost of physical quantum processing units presents a serious barrier to practical use. Moreover, of the four quantum devices tested, only IonQ’s Harmony achieves high fidelity with more than four qubits.”

Not surprisingly, the results showcase Terra/QMware’s performance – it is, after all, benchmarking its product against competitors. Pflitsch acknowledges this saying, “You’re right. [Still], we think [it’s] the first kind of holistic, comprehensive benchmarking exercise. In the future it should be third-party reviewed, but we thought the information is important. It gives some guidance to the industry in terms of hybrid classical-quantum platforms.”

In the press release for the study Terra Quantum said, “The goal of the study was to investigate how the prediction accuracy and training time of quantum neural networks can be improved by utilizing a hybrid quantum-classical approach. The benchmark found that the combination of simulated quantum processors and classical high-performance computers delivers the best performing, most cost-efficient, and robust approach.” (emphasis added)

It’s best to read the study directly. That said, the bolded section above captures the essence of the companies’ current pitch. Clever HPC architecture leveraging, for example, multi-threading, lots of memory, and novel memory access techniques, along with proprietary software has enabled Terra/QMware to build effective simulated qubits, says the company. No quantum device needed. The company argues that running quantum apps on its hybrid platform with simulated qubits outperforms common classical methods and outperforms today’s low-qubit-count QPUs.

Here’s a study excerpt: “The benchmark indicates an advantage in using the QMware basiq simulator for circuits with 2 to 26 qubits, AWS SV1 for 28-34 qubits, and QMware basiq for 36-40 qubits. Additionally, QPUs from four different vendors (IonQ, Ox- ford Quantum Circuits (OQC), IBM, and Rigetti) were benchmarked for runtime, accuracy, and cost. The results show QPUs could become time-competitive in a practical use case for circuits with 30 qubits or more. However, the current low fidelity attained by many of these systems precludes their application to industrial problems.”

Think of Terra/QMware’s current idea of hybrid quantum-classical computing as running hybrid applications on HPC hardware in which pieces of the application are run as they usually would be on HPC systems and other pieces of the application are run on simulated qubits, also on classical hardware. HPC is the engine underneath both. The results from each are integrated into a final solution. Terra/QMware’s current simulator can handle 40 qubits, according to the company.

It’s worth noting these are not quantum-inspired algorithms being run on a classical system, according to Florian Neukart, Terra’s chief product officer. “Quantum-inspired algorithms may, of course, also give some advantages. But that’s not the kind of algorithms that we mean when we talk about hybrid algorithms. Quantum-inspired algorithms take some paradigm from quantum computing or from quantum physics in general, say a superposition principle, and take that into a classical algorithm, for example, simulated annealing and try to do parallel tempering,” Nuekart said.

“Our algorithms are thorough hybrid quantum algorithms. Even though the QPU is simulated, it is as if it was a real qubit and that’s why there is a tremendous amount of hardware required, up to 12 terabytes of RAM to simulate 40 qubits. This is a real quantum system in the sense that it’s capable of everything that a 40-qubit quantum chip would be able to do, an error-corrected 40-qubit quantum chip. That’s the difference here,” said Neukart. “It’s very hard to jump from 40 to 41 qubits. We have a roadmap, even up to 42, but because you need exponentially more compute power to simulate all the quantum effects, it is hard to do.”

Pflitsch argues that the hybrid platform, leveraging simulated qubits, is delivering quantum advantage to customers now.

“An important strategic message is that hybrid quantum computing can unlock some of quantum’s potential today. We are, with our clients, not only dealing with toy problems, preparing them with the simulators for the future, but we are unlocking substantial business advantages with this approach, said Pflitsch.

He cited the example of an investment bank Terra is working with that has a €400 billion collateral portfolio.

“It’s a mission critical optimization within investment banking. Using data for the portfolio, we deployed our developed quantum algorithm on our machine and are able to get six pips (price-in-point or percentage interest point) advantage at full implementation. Six pips in relation to the portfolio translates into more than 200 million annual recurring cost savings. In the future it will be 60 pips if I put it on a better, native QPU but why wait for the 60 if we can today get the six pips. That’s what we focus on [and] it’s also a little bit of the Terra Quantum USP (unique selling proposition). We work with clients to [obtain] substantial business improvements using quantum hybrid quantum computing today,” said Pflitsch, who cited a few other companies Terra has worked with including Volkswagen (auto) and Uniper (energy) among them.

Long-term, Terra/QMware has broad ambitions. Pflitsch said little about the company’s superconducting qubit research but Neukart shared a broader description of the full platform and its ongoing integration with existing QPUs.

Currently, the overall QMware platform is broadly similar to other hybrid cloud-based approaches but with deeper integration, said Neukart. “Either you submit part of the problem to classical hardware or to quantum hardware, but you will always be limited to what one of these two can do. That’s where we differ. In that sense, we have integration, defined in the simplest way as web service integration. So that is where we already have already partners in place such as D-Wave, Xanadu, and Rigetti. But the most advanced stage of integration is currently in development that’s not been widely communicated publicly.

“What that means is we take our high-performance computing nodes, develop a specific hardware interface specific to whatever quantum chip we want to integrate, and then integrate the quantum chip in terms of a software representation into our operating system. So QMware really comes with the hardware and operating system, and Terra Quantum supplies software on top. But it’s the integration piece, the hardware interface and the operating system that is very different to what everyone else is doing. You could say that software sitting on top sees the hardware underneath as one big block of compute power,” said Neukart.

“Over time, as quantum chips become more powerful, we switch the simulator with a quantum chip and switch quantum chips with other quantum chips. [The] software on top doesn’t need to be touched; it becomes more powerful because the hardware underneath becomes more powerful. It allows us to easily parallelize. How we parallelize quantum processing units efficiently for machine learning, for example, is something that is integrated into the products that we have here. That is different. And these integrations, the advanced integration, are currently in the making. There’s nothing publicly announced yet, because we’re under NDA with them,” he said.

Interesting stuff. Stay tuned.