A System Architect’s Perspective on Quantum Computing: An Interview with Dr. Gokul Subramanian Ravi
Dr. Gokul Subramanian Ravi has been a career computer scientist. He is an assistant professor and active researcher in quantum computing at the University of Michigan. In this interview, we talked about the philosophy and technology of quantum computers. Dr. Ravi talked about quantum computers, its need, current state, architecture, and quantum advantage at length. While this interview is more technical than previous ones in quantum chats, it is very enjoyable and informative.
Mihir: Why do we need quantum computers? Classical computers are getting better every day. Can’t we just use classical computers for everything?
Dr. Ravi: That’s a good question. The classical computers’ capability is not increasing as fast as it used to be. We all have heard about Moore’s law failing. Thus, there is a fundamental need for new technology. We want more computing capabilities in any form, not specifically quantum computing. Today, computing and computers are the most fundamental driver of innovation. We want to keep pushing for new innovations. That reason motivates us towards emerging technologies, quantum computing being one of them.
Some problems are exponentially hard to solve. That means the computational resources required to solve such problems increase exponentially with the increase in problem size. Classical computers quickly reach their limitations in addressing this kind of problem. For example, to discover new medicines, you want to understand chemistry between two chemical molecules. You can’t mix thousands to chemicals together in the lab and study them, so computer simulations are done. Computational models for solving such problems represent each molecule as some numerical interaction and perform calculations to predict the molecular interactions. As the size of molecule increases, the numerical equations become exponentially complicated and soon reach the limits of classical computers. The reason for that is at the molecular level, when we are considering electrons; we cannot ignore some of the non-trivial forces, which we generally ignore in day to day calculations like gravity. With number of electrons, these forces become very large in numbers, hence the exponential growth of the problem. These are called quantum mechanical properties.
When Richard Feynman proposed quantum computing, the idea was that we needed a device that was able to simulate quantum mechanical properties and such a device would be quantum computing machine. Thus, quantum computer is specifically of interest for solving large-scale scientific problems in physics, chemistry etc. Other problems, like factoring, also have important application of quantum computing. If you are able to factor a number quickly, that has implication in security and cryptography. Factoring is a classical problem, not quantum, but there is a method that can solve factoring faster than a classical computer can. Quantum computer has a long way to do. However, in theory, there are quantum, classical as well as scientific problems that can be solved more efficiently using quantum computer than any classical computer.
Mihir: As you said, classical computers are reaching its capacity and no longer growing as fast as they were. As a result, we need new technologies to fill that gap and continue expanding our computational power. Is quantum computing one such new technology or are we calling a group of technologies quantum computing? Are we able to define quantum computer today?
Dr. Ravi: Again a very good question. In general, we would define quantum computer as a technology that is able to exploit quantum mechanical properties towards computing. Within that definition, all different technologies like supercomputing qubit, trapped ion qubit, neutral atoms, and photonic qubits are quantum technology. In their own way they all are exploiting quantum mechanics. If we are being very specific than you are absolutely right that quantum computing is an umbrella term. However, broadly they all fall within the same scope of exploitation of quantum mechanics.
Mihir: In my understanding, a problem has to be converted to a mathematical formula to make an algorithm that can be computed by a classical computer. Is that true for quantum computing also?
Dr. Ravi: I would say yes and no. I would approach this question in two different ways. Think of a problem which can be solved 90% on a classical computer and only last 10% needs a quantum computer because that last part is really exponentially hard. In classical computer, we would use an approximation and perhaps accept a 90% solution. We still need mathematical formula to reach that 90% solution and then improve beyond 90% using a quantum computer. We want to continue to use classical computer to go as far as we can, because quantum computer is always going to be an expensive resource.
Now the other question: is the quantum computing also based on a mathematical formula? I would argue, yes to some extent. Let’s take an example of a classical computer. In designing a complex machine learning algorithm, the algorithm would have complex metrics, its addition, multiplication and many complex mathematical operations. When coded onto a classical computer, a compiler would take that and through multiple steps ultimately pass down to transistors. Transistors would always work in a series of 0 and 1, no mathematical formula there. Thus, classical computer is formulas up to transistors and then it is just transistors’ natural property of 0 and 1.
Quantum computer is not much different. Let’s take example of chemistry. Let’s assume that we are trying to find energy of some chemical molecule, a common problem in chemistry. There are techniques like Jordan Wigner method, which converts fermionic (chemistry) form to the qubit form. There would be cleaning and optimization steps to remove non-important components from the molecular formula and properties. Finally, the qubit form is run on a quantum computer. If we assume there are twenty steps in calculating molecular energy, than nineteen of them are mathematical like cleaning, optimization, Jordan Wigner transformation and so on. Only the twentieth step is quantum computing, similar to going to the transistors in classical computing.
Mathematics and software gets less focus in quantum computing, because everybody is focused on qubits. Whereas in classical computing, we don’t think about transistors anymore.
Mihir: Let’s pivot now to system architecture. What is the simplest way to define system architecture irrespective of technology?
Dr. Ravi: Entire system is made up of multiple layers known as abstraction layers. One layer is an application like zoom or software doing chemistry calculations. Second layer is algorithm that application runs on. Then you have instruction layer like instruction set architecture which runs your device. To convert algorithm to instructions, you need a compiler. You may also have an operating system that is doing resource management. Another layer is micro architecture of the computer, which is how the computer is designed. This micro architecture has components like circuits and circuits are made up of transistors for classical bits or qubits.
System architecture is interactions between these different layers. Hardware architects focus on interactions between circuits, transistors and qubits like hardware components. Architects working at micro-architecture levels organize components within a processor. Other types of system architects deal with interaction between compiler and hardware, or compiler and algorithm, or stacking servers to build complex super-computing architecture. System architect is a broadly defined term for a group of experts working anywhere among different layers of hardware and software and they understand the pathway from application to technology. It is a complex pathway and system architects usually work on only a subset of different layers.
Mihir: How has the role of system architect evolved over the year?
Dr. Ravi: Yes, the role has definitely changed over the years. That change has come based on the needs. During the seventies, there were so many opportunities and needs in a single layer of the stack that a person can focus on being expert of just one layer like on micro-architecture or compiler. Early 2000s, computers started to reach limits of computational power within a single core and multi-core systems became a norm. That prompted change in the role of some system architects. They asked questions about parallelization of processes, dependencies between applications and different cores and other questions that system architects did not think about before. Because the capacity of processors was not increasing rapidly, the focus shifted to building accelerators. Again that had an impact on role of system architects. The architects needed to look at multiple layers from application to processors, but they were focusing on just one application. Earlier system architect’s role was broad within a layer or two. Modern system architect’s role has become deeper than broader.
Mihir: While systems architecture was evolving for classical computing we had opportunities to try and fail. Now that we have all these knowledge about computing, we have to use our knowledge in quantum computing. We do not have enough opportunity to try new things and fail, isn’t it?
Dr. Ravi: Again, a very good question. On one hand it has been a huge positive that we have learned to build a full stack in classical computing and we can apply that knowledge to quantum computing. For example, IBM has been at the forefront of building system architecture for classical computing; it is applying that knowledge to the quantum computing and doing very well.
On the other hand some of the strategies and habits that work in classical computing may not work in quantum computing. In emerging technology you can’t start with being broadly expert in one layer like how classical computing started. We have to be flexible. As other layers are evolving, system architect in quantum computing needs more depth and flexibility in their knowledge and approach.
