This past week has been particularly active with new quantum product and feature announcements. This is partially due to everyone wanted to get these releases out before people start going on vacation for the holidays and also because some of these company wanted to announce these at the recent Q2B Conference. In this article we will provide summaries of all the announcements we saw and may publish more detailed analyses of specific announcements in the coming weeks.

Amazon AWS – TN1 High Performance Tensor Network Simulator
Amazon AWS released a high performance quantum circuit simulator using tensor network algorithms. These simulators can be significantly higher in performance than other simulators if they can leverage circuits with an inherent structure. The TN1 can simulate quantum circuits of up to 50 qubits and the graph below shows a performance comparison of the TN1 versus the SV1 state vector simulator. Additional information about the TN1 can be found on the AWS website here.

Performance Comparison of AWS SV1 versus the TN1 Quantum Circuit Simulators, Credit: Amazon

D-Wave – Qiskit Plug-In for Enabling Comparisons of Optimization Performance
D-Wave’s machines are suited for solving problems that can be expressed in a quadratic unconstrained binary optimization (QUBO) format. These types of optimization problems have many potential real world uses. But the question remains on whether the D-Wave processors can solve these problems better than the current gate based machines. To help customers answer this question, D-Wave has released an IBM Qiskit plug-in that can take optimization problems programmed as problems using Qiskit as a front end and convert them to a format that can run on the D-Wave Advantage quantum annealing processor. The problems can also be run on IBM’s Q systems as well as the growing number of other platforms that can accept inputs from Qiskit and compare the results.  For more on this new plug-in, you can read the news release on the D-Wave website here and also download this plug-in located on GitHub here.

Google – Quantum Simulations Using Their Tensor Processing Units (TPU)
Google had previously designed their own special purpose ASIC called a Tensor Processing Unit (TPU) for use in special purpose machine learning algorithms. This unit is much faster than general purpose CPUs or even GPUs for these applications. They have now programmed this device to do very high performance quantum simulations. It can currently handle simulations of up to 32 qubits and are working to expand it to 36 or 40 qubits. As can be seen in the chart below, a TPU powered simulation is over two orders of magnitude faster than Google’s standard Cirq simulator and one order of magnitude faster than an optimized Qsim software simulator that Google has developed. Because of the unique computer architecture inside the TPU, this device will be able to provide some of the highest performance classical simulations of quantum algorithms than anything else around. The TPU powered simulator is currently in alpha testing and will be more generally available in 2021.

Google TPU Powered Simulation Performance Comparison, Credit: Google

IBM – Steps to Improve Hybrid Classical/Quantum Computing
A very large number of quantum algorithms being investigate today utilize hybrid classical/quantum solution to achieve the best result. Certain algorithms, such as QAOA and VQE, may require the two computers to pass results back and forth thousands of times before they arrive at their final solutions. magnitude and make a solution unworkable.

Whereas reduce delays from seconds to milliseconds may be acceptable for the hybrid algorithms like QAOA and VQE, there are other situations where classical control needs to be performed within the coherence time of a qubit which is currently in the 100 microsecond range. These situations include implementation of a conditional reset capability in the processor, implementation of some error correction algorithms, or possible implementation of what we call a Quantum IF statement within a quantum program.

To implement these capability, IBM will be developing both hardware and software technologies to perform these functions. They have classified the improvements into two buckets which they call near-time classical for the hybrid algorithms and real-time classical for the microsecond level qubit control.  The chart below shows pictorial how they see the evolution of the systems.

Comparison of Current versus Future IBM Hybrid Classical/Quantum Architectures, Credit: IBM

This chart below shows how they view that the execution model will also change for the near-time classical portion of the technology. By taking the multiple iterations out of the queue, significant time savings will be achieved. Also, since the classical processor and the quantum processor will be in close proximity to one another, there will also be savings due to elimination of transit delays. In an example, of the savings for using this new technique, an IBM engineer estimated at the Q2B conference that the simulation time of a Lithium Hydride (LiH) molecule could be reduced from 111 days to 16 hours.

Comparison of Current versus Future IBM Classical/Quantum Execution Models, Credit: IBM

Although a lot of engineering still needs to be performed to fully implement these capabilities, IBM has taken the first steps by creating a proposal for an enhanced version of the OpenQASM language that they call OpenQASM3. These concepts are not entirely new. Other quantum teams have previously implemented similar capabilities in their systems. For example, Rigetti has implemented something that they call co-location which has a high performance classical processor sitting inside their quantum data center with the quantum processor. And Honeywell has implemented a capability similar to what IBM calls real-time classical with a capability called mid-circuit measurement.

So IBM still has a lot of development required to make these concepts a reality, but we do expect to start seeing them in the future machines described in the roadmap that IBM released in September. For more on these capabilities, you can view IBM’s blog post describing them here.  For more on the OpenQASM3, you can view another IBM post on Medium here.

QuiX – A 12 Mode Universal Photonic Processor for Quantum Information
QuiX’s photonic processor uses a qumode as the basic unit of information similar to Xanadu’s photonic implementation. It is a low-loss, 12-mode fully tunable linear interferometer with all-to-all coupling based on stoichiometric silicon nitride waveguides. It is arranged in a 12×12 configuration and can be programmed using the thermo-optic effect using heat to steer the path of light. The device is compatible with all photon sources and is available now.

QuiX Photonic Processor, Credit: QuiX

For more information about QuiX’ photonic processor, you can view a news release on their website here, an arXiv preprint that describes their technology in more detail here, a YouTube video describing their 12 mode device here, and a Medium post that explains the difference between qumodes and qubits here.

Rigetti – Language Extension to Quil Called Quil-T for User Pulse Level Control of Qubits
Deeper level control of the control pulses in a quantum computer can provide enhanced capabilities for a user’s program. This includes improving gate fidelity, creating new gates types to reduce circuit depth, and enforcing precise program timing. The capability has been available with a module called OpenPulse in IBM’s Qiskit. Up until now, Rigetti users were not able to control the qubits down to this level and could only use the standard gates and pulse timings and sequences built into Quil, but with this new language extension users will now be able to take advantage of this new capability. It has been implemented on Rigetti’s Aspen-8 32-qubit processor and will be available in the coming months to users who access that machine through Amazon’s Braket service. Additional information about Quil-T is available in a Rigetti blog posting on Medium. In addition, Q-CTRL, a quantum software company that specializes in implementing these type of qubit controls, has been working with Quil-T and has published their own blog article showing some examples that you can view here.

Xanadu – Pennylane Software Package is Now Available on Amazon Braket
Pennylane is Xanadu’s open-source software for providing differentiable quantum computing capabilities. It is designed for the development of variational quantum algorithms and quantum machine learning (QML) applications. It has previously included support for IBM, Google, Rigetti, and AQT hardware platforms and has added an integration to work with Amazon’s Braket service. Braket currently supports the Rigetti and IonQ gate-level platforms but has indicated they plan on adding more in the future. Braket also supports the D-Wave quantum annealing processor, but we do not think Pennylane will work with that platform. Additional information about this new support can be found in Xanadu’s news release which you can find here and a post on the Amazon AWS Braket website here.

December 11, 2020