IBM held its annual IBM Quantum Summit today and made seven significant announcements about the status and plans for its quantum computing activities. Although there was a lot of press hype about the 127 qubit Eagle processor, many of the other announcements were ignored and some may even be more significant than the Eagle itself. Here are some of the key things they mentioned.
127 Qubit Eagle Processor
IBM has had this processor on their public roadmap since September 2020 and has frequently mentioned it this year. So this should be no surprise to anyone who has been following the industry. We would point out that the IBM Eagle release is the first of several of what I call “adolescent-sized” (larger than 75 qubits) quantum computers that will be introduced in the next few months. Other companies that have discussed near-term roadmaps for machines of this size include Rigetti, ColdQuanta, Atom Computing, Pasqal, and perhaps a few others. A key characteristic of computers of this size is that their operation can no longer be simulated with a classical computer because the computational resources required double every time another qubit is added. So classical simulation of these sized machines becomes an intractable problem.
This machine includes all the innovations that IBM put into their previous Falcon class processors including their heavy hex topology, laser annealing to fine-tune the qubits, multiplexed readouts to reduce the number of cables, and added 3D packaging for better routing of the connections. Their next-generation Osprey 433 qubit machine scheduled for 2022 will have all these innovations and will add scalable I/O with new flex cables, improved control electronics, and a new packaging called IBM System Two. (See below)
The machine is not yet available to their customers but is anticipated to start becoming available to their premium customers in December. The machine is listed as the IBM_Washington on the web page that lists their machines, but key parameters, such as Quantum Volume, have not been released yet. We expect that IBM is still working on calibrating and tuning the machine to achieve the best possible qubit quality. But this can be a challenge because adding more qubits can sometimes detract from the overall qubit quality levels. For example, IBM’s IBMQ_Falcon machine with 27 qubits shows a Quantum Volume of 128, yet the IBMQ_Brooklyn machine with 65 qubits only shows a Quantum Volume of 32.
IBM Quantum System Two
One thing that was a surprise to us was IBM’s announcement they will be adopting a new physical form factor and packaging they are calling IBM Quantum System Two. We expected that their IBM System One form factor which was only introduced in 2019 would be sufficient to support all of their designs up to their 1121 qubit Condor design scheduled for 2023. However, they have decided that the Eagle will be the last generation to use the IBM System One and their future 433 Osprey and 1121 Condor will use this new IBM Quantum System Two form factor. IBM is developing this in partnership with dilution refrigerator manufacturer Bluefors and a key motivation for adopting this is to provide modularity. One concept that IBM is exploring is a future quantum data center that contains multiple quantum computers that are connected together. The IBM Quantum System Two form factor is arranged in a hexagonal form factor and this may make it much easier to locate multiple quantum computers next to each other. It will also provide for a new generation of scalable qubit control electronics along with higher-density cryogenic components and cabling.
Improved Qubit Coherence Times and Gate Fidelities
As mentioned above, improving the number of qubits is not enough. IBM is working to improve the quality levels of the qubits too. Currently, the best machines they have online now, like the IBMQ_Kolkata or the IBMQ_Montreal have T1 coherence times of 100 microseconds and Average CNOT gate errors of about 0.01. They described in the conference recent experiments that have increased the coherence times to 300 – 600 microseconds and CNOT gate errors at 0.001. These would represent very significant improvements, but we are not sure if all this goodness has made its way into the 127 qubit IBM_Washington machine yet. We should find out in a few weeks when IBM releases some of these quality metrics for that machine. Longer-term, they are shooting to achieve coherence times about 1 millisecond (1000 microseconds) and gate error levels at about 0.0001 in 2024 for the Condor class machines.
IBM has recently been emphasizing circuit execution speed and reiterated that this was a key focus for them. We’ve recently described many of their activities in this area in our articles discussing the CLOPS (Circuit Layer Operations per Second) measure as well as Qiskit Runtime to streamline the interfaces between the quantum and classical computer. IBM announced that they have completed the implementation of the Qiskit Runtime on all their currently active machines. You can view our previous articles describing these here and here.
Circuit Knitting and Dynamic Circuits
IBM has also been exploring ways that they can leverage the combination of a classical computer working with a quantum computer to provide a synergistic improvement in the computer’s capabilities. One technique they call Circuit Knitting involves taking a large quantum circuit, splitting it into pieces, running the pieces separately, and then having the classical computer combine the pieces together to get the answer. Although this may not work for all quantum circuits, this method may be able to allow a user to reduce either the number of qubits needed or the levels needed to run their program and it may allow the user to fit a larger problem into a smaller machine. Another use would be to implement various forms of error mitigation by doing multiple quantum runs and having the classical computer combine them together to provide a more accurate result. Both of those techniques provide a tradeoff by providing either a larger scale or higher accuracy at the expense of longer runtimes. So that is another reason why the improved speed discussed in the last section is important.
Another innovation that IBM is planning for 2022, is called dynamic circuits. Honeywell introduced something similar in 2020 with a capability they call mid-circuit measurement. The idea is that a quantum program can measure an individual qubit and take a branch in the code based upon the outcome of the measurement. Besides providing a branching capability, a dynamic circuit can also enable qubit reuse. A qubit can be measured, reset, and used for something else later on in the calculation. This capability can also reduce the number of qubits needed to implement a calculation. IBM mentioned they were able to simulate an H2O molecule with only 5 qubits using this technique.
Serverless Computing and Integration with the IBM Cloud
So far, the IBM Quantum Cloud has been separate from the standard IBM Cloud. But this will change in 2022. They will do an integration so their standard IBM Cloud customers can access the IBM Quantum Cloud with their IBM Cloud account. A limited beta will start in January with a full beta in March and general availability in the summer of 2022. IBM has a goal of frictionless computing and they are taking steps to make it as easy as possible for a user to take advantage of their capability. They are aiming to provide customers with seamless scaling where they can use the capabilities without worrying about configuring which system they need to use and with customers only paying for the actual consumption they actually use.
More Systems Installed Externally
Earlier this year, IBM installed their first two systems outside of the United States in Germany and Japan. At the Quantum Summit, they announced they will also install a system at Yonsei University in Seoul, Korea. In addition, IBM is scheduled to install an IBM Quantum System One at the Cleveland Clinic but it was also mentioned at the conference they will also install an IBM Quantum System Two based machine at the Cleveland Clinic sometime in the future.
Perhaps a summary of IBM’s quantum strategy can be captured with two mathematical formulas. The first is that Performance = Scale + Quality + Speed. And the second is Value = Performance + Capabilities + Frictionless. So it is important to understand that IBM is working on many different dimensions to continually improve their quantum efforts. One comment made by Jay Gambetta, IBM Fellow and Quantum Vice-President is a belief that as long as IBM continues to meet their roadmap, we will start seeing end users achieve a quantum advantage by 2023 with their Condor class machines.
For more about the IBM Quantum Summit and their recent announcements, you can view a press release announcing the 127 qubit Eagle processor here, a blog article about the Eagle and the IBM Quantum System Two here, a second blog article on Quantum Serverless here, and another press release announcing IBM’s collaboration with Yonsei University here.
November 16, 2021