Subtitles section Play video Print subtitles [MUSIC PLAYING] MARISSA GIUSTINA: Quantum computing-- it's been all over the news lately, often accompanied by an air of mystique or an assortment of fantastic promises. But what does "quantum" mean in the context of computer hardware? What distinguishes a quantum computer from a regular one? What does a quantum computer look like? How do we build it? My name is Marissa Giustina, and I'm a research scientist in the Google AI Quantum Hardware Lab. I'd like to unpack those questions. And hopefully, in about five minutes, the term "quantum computer" will have just a little more meaning for you. We're working to build devices that we can interact with. That is, devices we can control and read out, which behave reliably according to a simple quantum model. In other words, we're building quantum computing hardware. Quantum hardware can be used as a tool for approaching certain kinds of computational problems. So our ongoing efforts are both to develop the hardware and to develop algorithms that leverage this hardware. Let's start with the first question. What does it mean for hardware to be quantum? For that, we'll talk for a moment about quantum mechanics. A model is the physicist's tool to make predictions about what will happen when we put the universe into a certain configuration and poke it in a certain way. For example, if you'd never built a skyscraper before, you might make a Lego version before building it full scale. That's a model. Models can also be expressed in the language of mathematics. The most fundamental model of nature we know was developed in the early 20th century and is known as quantum mechanics. The word "mechanics" refers to the mechanisms by which things happen. The word "quantum" refers to discrete quantities of energy or some other physical quantity. Within quantum mechanics, energy comes in packets, sometimes called photons. And you cannot have fractional packets. So what's a quantum object? People sometimes think of a quantum object as being tiny and a quantum leap as being large. However the word "quantum" doesn't dictate an object's size. Actually, a quantum object is one that relates in a well-defined way to a single quantum of energy. For instance, the photon I mentioned before is a quantum object. A photon is a single particle of energy. Similarly, atoms are quantum objects. An electron flying around an atomic nucleus may be excited into a higher orbit only by a particular quantum of energy. There is no halfway point between the lower orbit and the upper orbit. If the wrong energy is provided, there simply isn't a corresponding orbit for the electron to land in. In a nutshell, a quantum object is one whose observable behavior reflects that nature only offers energy in discrete packets. Now onto the next question. What differentiates quantum computing hardware from a regular computer? In essence, quantum hardware lives in a richer world than its conventional counterpart. Let's consider a simple, abstract, quantum object, which is entirely described by the fact that it can be in one of two different energy levels. Let's call those levels 0 and 1. You can interpret those brackets around the 0 to mean this is a quantum energy level called "0." And likewise for the "1." Here, for example is a quantum energy state named "psi." Recall the classical bit of information, a switch that can take one of two values-- 0 and 1. Because of the apparent similarity between our quantum object and that classical bit of information, we call this quantum analog a quantum bit, or qubit. One peculiar feature about quantum mechanics is the existence of superpositions. A superposition is like a special mixture of the energy levels 0 and 1, where the weight of each energy level is given by complex constants C0 and C1. If we measure the energy of our qubit, we will sometimes observe 0, and sometimes 1, where the value of sometimes is given by the constants. An individual measurement will yield an outcome of 0 or 1. There are no other options. But before the measurement occurs, we know at most the chances of getting a 0 or a 1. We can't know the actual outcome for sure until we measure it. Therefore, when we want to talk about the energy state of the qubit before we've made the measurement, we use this superposition to represent that the qubit hasn't decided yet which outcome to display, even though the chances of getting each outcome are fixed. Now, even admitting that this superposition business is a little unusual. We can accept that it's easy enough to represent one qubit. We just wrote it down right there. Thinking about more qubits gets increasingly difficult. Suppose we add a second qubit. If these were conventional switches, we could think about each switch independently. But qubits are different. Just as one qubit can be in a superposition state, two qubits can share a superposition state, where, for instance, the measurement outcome is unknown, but will certainly be the same for both objects-- or opposite for both objects. For example, here's a state where a blue qubit and a yellow qubit are together in a superposition state. Here, they're correlated to each other. Before the measurement, it cannot be known whether the blue qubit will turn up 0 or 1. But a measurement of both qubits will certainly always give the same answer for each. Similarly, in this case, measuring the blue and yellow qubits will always give opposite outcomes. This means that in order to fully describe two qubits, we need to consider C's for all possible measurement outcomes we could see. To describe three qubits, we need eight C's. Describing four qubits takes 16 C's, and so on. Each time we add another qubit, it takes twice as much information to describe the whole pile of them. That is the crux of what differentiates quantum hardware. The quantum system lives in a richer space, so that representing n qubits with a classical computer requires 2 to the n numbers. But does this mean that a quantum memory with 100 qubits corresponds to a conventional memory with 2 to the 100th bits? Not so fast. Quantum hardware is very effective at encoding and processing certain kinds of information. But it cannot efficiently mimic many useful aspects of its classical counterpart. When we say that a picture is worth 1,000 words, we don't abolish words entirely in favor of pictures. Adding quantum hardware to our modern computing capabilities would be like adding pictures to a communication strategy that, up to now, used only words. So what does quantum hardware do well? The exponentially growing complexity of quantum systems also gives a clue about where quantum hardware could be useful. In the fields of chemistry and materials development, simulation of molecules could be a powerful technique to learn about the properties of a new molecule before fully synthesizing it in the lab. However, our ability to simulate chemistry on computers is limited. At its heart, chemistry is an application of quantum mechanics. And each electron we add to a model doubles the number of parameters, crippling computers with expensive calculations already for very small molecules. Suppose instead that we could build chemistry models out of a quantum Lego set. Then the model would be built with the same physics that governs the system being modeled.