In late 2020, IBM released its first long-term quantum roadmap, showing how IBM’s quantum architecture, hardware and qubit count would change over the next few years. IBM plans on evolving its present-day small-scale, noisy quantum computers to a near-term intermediate 1121-qubit machine named Condor. Once perfected, Condor will become the future building block of a larger fault-tolerant quantum computer with millions of qubits.

Qubits represent the fundamental unit of information in quantum computers. Unlike classical computing bits, which can only represent either a one or a zero, qubits can also be a one or a zero or a superposition of both values. Superposition is a fundamental feature of quantum mechanics that plays an essential role in quantum computing.

Last week, IBM released a new and more descriptive technology roadmap. It overlays an expanded timeline of future applications, new Qiskit software and developer capabilities on top of the earlier 2020 hardware roadmap.

According to Jay Gambetta, *IBM* Fellow and Vice President Quantum Computing, IBM recognized more future plans were needed in its roadmap. “Ultimately software is really tied to the hardware. What I wanted to do this year was to put some context around where we see the software going, and then bring it together with more of an application focus for the user.” Gambetta went on to say he believes quantum computing will eventually be able to solve “big problems” in the areas of natural sciences, optimization, finance and machine learning.

Quantum solutions to problems in these four areas will ultimately touch and influence almost every facet of our lives. The first working 2-qubit quantum computer was announced in 1998. Since then, quantum scientists have dreamed of building a universal fault-tolerant quantum computer with millions of qubits. However, for many years, some scientists didn’t believe it could be done.

**New IBM 2021 development roadmap**

IBM’s hardware and qubit counts remain unchanged from its first 2020 roadmap. However, for 2021 and beyond, IBM will focus its efforts on developing software that allows circuits to run faster and makes it easier for developers and industry specialists to use quantum. Moreover, these software improvements will happen in a future environment where integrated classical computers and quantum computers will provide a seamless quantum solution. After a careful review, it is clear that IBM is building a complete software ecosystem around users of its quantum cloud. Gambetta believes that for technology to be adopted, IBM needs to make it as frictionless as possible. Moreover, he believes developers shouldn’t have to learn new languages. Gambetta says quantum programming must be integrated into developers’ existing code and easily called with a cloud quantum API or service for new quantum technology to be successful.

**Software tailored to developers**

In 2016, IBM provided the world’s first cloud access to a superconducting quantum processor with five qubits. Almost immediately after launching the system, papers were published based on research performed on the system. Since then, quantum researchers have made significant contributions to the evolution of quantum computing.

Today, IBM has over 20 quantum computers available on the cloud, with over half offering free access. Usage on IBM’s quantum cloud is staggering. Over 1.3 billion quantum circuits are run daily, and democratized cloud access for researchers has resulted in over 300 technical papers. From the time IBM’s first quantum computer became available on the cloud until now, there have been over 700 billion quantum cloud executions.

According to the roadmap, IBM is creating a user-friendly software approach for developers which will facilitate access to future quantum services. The company will be customizing access to its quantum hardware based on specific interests, needs and existing coding environment of developers. Robert Sutor, Vice President of IBM Quantum Ecosystem Development, said, “We have laid out a software approach heavily oriented towards developers. We feel strongly that a healthy user base will also be a guiding force that will help shape the future technical direction of quantum devices.”

Qiskit is IBM’s open-source quantum programming framework that allows researchers and developers to program quantum computers and classical simulators. IBM’s primary goal is to increase its hardware capacity while making its quantum programs simple to use for the largest number and greatest variety of developers possible. Each type of developer has its own separate and distinct needs.

The following developer descriptions were derived from an earlier IBM paper and edited for clarity. IBM plans on creating a “frictionless” software ecosystem for each type of developer, offering access in a form familiar to them. IBM also intends on providing developers access to data associated with that work level, such as coherence times, qubit frequencies, crosstalk and error rates for calibrated quantum gates and operations.

- Model developer – A quantum developer is usually an industry expert who is primarily interested in seeing how an application in its field works in a quantum environment. For example, rather than worrying about the technical circuit and device details, a quantum developer in the field of finance might be more interested in a straightforward comparison of results between a Monte Carlo simulation on a classical computer versus the outcome of
*Monte Carlo*sampling using quantum amplitude estimation. Their objectives are to run circuits for an application and receive a result as quickly as possible. - Algorithm developer – Typically, an algorithm developer is a quantum information scientist or software engineer who completely understands a quantum circuit and how it runs on existing noisy quantum computers. There are many ways to implement a quantum circuit. Finding the optimal solution is computationally complex and is an important research topic. Developers are interested in exploring error-correction primitives, such as parity checks and conditional operations that depend on multi-qubit measurements.
- Kernel developer – The kernel developer is likely to be a quantum physicist or hardware engineer who understands the underlying device physics. They are interested in technical details, such as optimal control techniques, novel pulse-shaping approaches, techniques to quantify the underlying system, Hamiltonian, and error mitigation methods. These users require the ability to examine device-level properties of the system, e.g., control over the frequency, timing, pulse shapes and measurement integration kernels.

**Future IBM software developments**

Circuits provide instructions for quantum computers. In the early stages of quantum computing, it made sense for IBM to focus optimization efforts on improving circuit capacity and circuit quality. Leveraging these previous circuit improvements, IBM will be releasing a feature called Qiskit Runtime for kernel developers sometime in 2021. Runtime will provide faster circuits and allow programs to be stored and shared with other developers.

For example, running a chemistry algorithm today is a complicated process. Before executing any circuits, you must pick the plot points, choose the error mitigation and classical quantum optimization algorithms, then recast the problem to fit the quantum machine. Lastly, you need to consider how many shots are needed. Continuing this full loop allows the developer to do calculations on their classical computer using data from the quantum computer.

IBM plans to simplify the process by putting these steps together and then executing them close to the quantum processor. Lithium Hydride is a relatively small molecule that IBM uses as an example to illustrate runtime speedup. Current simulation of the molecule can require up to 100 days. Runtime will shorten the simulation to a day or two.

**2021 Mid-Circuit Measurement and Reset**

Measuring a qubit causes its superposition to collapse, revealing its state to be a one or a zero. That is why current measurements occur at the end of a quantum circuit. However, IBM has already introduced a new feature called mid-circuit measurement and reset (MCMR). MCMR allows measurement of a qubit at any point in the circuit and triggers other actions. Regardless of its measured state, the qubit is reset to 1 so that it becomes a known state, which allows it to be reused, making more efficient use of resources. MCMR can also be performed multiple times in a circuit.

**2022 – Dynamic Circuits**

IBM has prototyped “smart circuits” called Dynamic Circuits that will be available in 2022. Dynamic circuits are circuits in which future states depend on outcomes of measurements that happen during the circuit. Dynamic circuits will allow branching actions such as the use of real-time classical processing to take place based on conditions within an existing circuit. Dynamic circuits can be useful for demonstrations of dynamic error correction, classical logic, developer assertions, and zero state preparations. IBM expects Dynamic circuits to be widely used and contribute to creating a wider pool of circuits available to developers.

As shown in the above circuit diagrams, dynamic circuits using MCMR can also be used for a fundamental quantum algorithm called quantum phase estimation (QPE). Many algorithms use QPE because it has the potential to provide logarithmic speedup. Phase estimation is also an important part of period finding to factor numbers in Shor’s Algorithm (one of the most famous algorithms in quantum computing). Unfortunately, running quantum phase estimation requires many resources and many shots to obtain an accurate answer.

The above IBM illustration compares two methods of phase estimation: post-processing vs. real-time using dynamic circuits. The basic question for this scenario is which solution needs the least number of resources to obtain the answer with the specified accuracy? IBM researchers recently ran a version of the quantum phase estimation algorithm (iterative quantum phase estimation) with dynamic circuits. The researchers proved dynamic circuits took fewer resources than other methods. Once this feature becomes available, IBM believes dynamic circuits will become an essential software tool for kernel developers. Moreover, its use should produce many papers that advance its future use.

**2023-2026 **

*Hardware*

According to the roadmap, a significant hardware milestone will occur in 2023. That’s when IBM plans to introduce its 1121-qubit Condor quantum processor. The Condor will be preceded in 2021 by a 127-qubit Eagle processor and in 2022 by a 433-qubit Osprey processor. Even though 1121 qubits may sound like a monster by today’s standards, we will need a machine that is thousands of times larger to fulfill quantum computing’s true potential. Even so, the Condor should be able to do some useful work, perhaps even achieve quantum advantage for limited applications. This machine should allow IBM to make significant progress with error correction. The Condor will also help researchers develop and optimize a large qubit architecture to prepare for the million-qubit machine. Beyond 2026, IBM envisions having advanced control electronics and software that seamlessly integrate classical HPC and a fault-tolerant quantum computers with millions of qubits.

*Software*

IBM will begin releasing circuit libraries to provide kernel developers with tools to investigate algorithms that use large qubit hardware. According to the roadmap, advanced versions of dynamic circuits will be segmented, then reconstructed into larger circuits tailored to specific needs. Later, frequently run circuits can be used to create groups of pre-built quantum runtimes. These runtimes can be customized for specific industries, then called by APIs using common development frameworks. By this time, IBM believes its 2021’s “frictionless” strategy will have attracted enough kernel and algorithm developers to produce a large body of usable research and algorithms. Both model developers and enterprise developers will benefit from this research, enabling them to explore quantum computing models without needing academic training in quantum physics.

*Analyst notes:*

- There are about six common technologies in use for building quantum computers. The most common qubit technologies today are superconducting, trapped-ions, and photonics. IBM uses superconducting transmon qubits, a name derived
*“transmission-line shunted plasma oscillation*qubit*.”*It is not yet clear which technology will be declared the best technology to use for quantum computing. It’s possible that the best technology is yet to be discovered. - Google also published a long term “roadmap” with just a few incremental superconducting hardware jumps leading to a multi-million-qubit quantum computer. The roadmap has minimal hardware information and no details about how software or applications will evolve with hardware improvements and growth.
- Honeywell Quantum Solutions has been using mid-circuit measurement and reset in its trapped-ion quantum computer since its Model H0 release in March 2020.
- To illustrate today’s state of quantum computing development, imagine the difference in a laptop’s capabilities compared to the computational power of IBM’s Summit supercomputer. That is about the same difference that will exist between today’s small quantum computers and a future fault-tolerant quantum machine. Such a machine would contain millions of qubits capable of computing the prime factors of a large integer and solving other important problems.
- Unfortunately, there is no quantum computer architecture in use today that can scale up to a large fault-tolerant size by simply adding more components. We have identified many engineering and physics issues that we don’t know how to solve. Likely there are other future problems we have yet to identify. In other words, we don’t know what we don’t know.
- Current quantum computers are susceptible to computation errors created by its environment and noise from internal components and qubits. Compared to classical computers that rarely have an error in trillions of calculations, today’s quantum computers experience an error every few hundred calculations. Almost every gate-based quantum company and many universities are conducting error-correcting research. However, no current software or hardware design will handle errors at a level necessary for million-qubit machines.
- Error correction is a significant area of quantum research. For IBM to attain its hardware milestones in 2025 and beyond, it must develop a workable error correction solution.
- Because of media hype and misinformation, many people believe that large-scale quantum computers are currently in use or just around the corner. The reality is that today’s quantum computers have limited computing power and are still in the prototype stage. A quantum computer that can solve problems more efficiently than a classical computer is at least 3-7 years away.
- Most quantum computing articles focus on the number of qubits. Qubit counts are front and center here, but only because qubits are the common denominator in almost every roadmap. Qubits are important, but they must be high-quality qubits. However, scaling up qubits to huge numbers depends on many factors, including the quantum computer’s technology, architecture, physics and the ability to fine-tune and engineer hardware components.
- IBM still considers quantum volume to be a valid measurement and a useful tool for engineering high-quality quantum circuits. Circuit quality translates into increased capacity because faster circuits can perform more quantum computations.