Whenever a new smartphone or home computer is released, the most discussed topic is always the computing performance of the central processing unit (CPU) and the problem of heat dissipation. We may not know the specific computing architecture inside the CPU of electronic devices, but we can truly feel the huge changes brought by the improvement of CPU computing power to our lives and work.
In fact, from the scientific calculators we often used in middle school, to the laptops necessary for daily office now, these electronic devices with data processing capabilities can be collectively referred to as electronic computers. So, on the basis of classical computers, are there faster and stronger computers?
(Image source: Veer Gallery)
A wonderful world of "0" and "1"
Every time we press a mobile phone key or computer keyboard, these characters or information first need to be converted into an encoding that the computer can handle: a permutation of 0 or 1.
For example, according to the character encoding of ASCII (American Standard Code for Information Interchange), the letter K is encoded as "01001011." After the calculation and processing of the electronic computer CPU, a string of 0 or 1 encoded information can be converted into letters or pictures that we know to present in front of us.
In the CPU of the electronic computer, the electronic transistor can be used as the basic unit of 0 or 1 for numerical calculation, at this time the path of the electronic transistor can represent the value 1, the open circuit represents the value 0, and when millions of electronic transistors and other electronic components are packaged on a small semiconductor wafer through the VLSI process, this integrated circuit with data processing capabilities and micro-encapsulation is generally referred to as "chip".
(Image source: Veer Gallery)
Not enough 16 cores? Then more "billion" points!
However, with the increasing number of data processing tasks in life and work, it is often difficult to meet the needs of the computing power of a single CPU core on electronic devices, and we can integrate multiple computing cores on the CPU to ensure that multiple cores can process data tasks independently at the same time. Today's common business computers generally use 8-core CPUs, and even some workstations used to process large computing tasks will have 16-core CPUs.
But even such a high-performance computer is difficult to meet more and more complex computing needs, taking our common weather forecasting as an example, the computer needs to discretize the atmosphere of a specific area into a grid for numerical simulation, and if you want to achieve the accuracy of the next 3 days to achieve more than 90% of the weather forecast, you need up to tens of billions of floating point operations, if ordinary commercial computers to calculate, it takes at least half a month.
In addition, urban intelligent transportation and online cloud computing supported by big data centers require exponential data processing capabilities, so people began to interconnect millions of CPUs to work together to build "supercomputers" that can calculate in parallel.
(Image source: Veer Gallery)
In fact, "supercomputer" is not a single computer with supercomputing power, but an abbreviation for a cluster of supercomputers. That is to say, each node of the supercomputer cluster is an independent computer, and its "super" lies in its own unique node interconnection structure, so that thousands of CPUs on all nodes can be scheduled at the same time, and there are generally dozens of physical cores in each CPU, thus having exponential data processing capabilities.
However, the "supercomputer" is not an all-round player, it only has powerful processing power for algorithm problems that can be parallelized, but it cannot accelerate the processing of data tasks for serial computing, in addition, the task scheduling strategy and compiler optimization of the "supercomputer" operating system also greatly affect its own performance.
The serial/parallel computing task here can be understood in layman's terms: a pile of earthwork originally required 1 person to move in 10 hours, and 10 people could also be arranged to move it in 1 hour. However, if one person is required to spend 10 hours digging a well that can only accommodate one person, it is impossible to arrange 10 people to complete it in one hour, which is a serial computing task that is difficult for "supercomputers" to handle efficiently.
Therefore, the "supercomputer" is not simply a stack of CPU and computing cores, and its accelerated computing capacity not only relies on the highly collaborative interconnection between CPU cores, but also optimizes the algorithm order of data processing tasks to fully call the computing potential of each CPU.
(Image source: Veer Gallery)
The well-known "Sunway Taihu Light" supercomputer has a total of 40,960 CPUs, and its peak computing speed reaches 1.254 billion billion times per second, and the continuous computing speed reaches 930 million times per second. The "Sunway Taihu Light" supercomputer helped the teams of Tsinghua University, Beijing Normal University and the Chinese Academy of Sciences to complete the "global atmospheric non-static cloud resolution simulation" and realized the global 10-kilometer high-resolution numerical simulation of the earth system, which further enhanced the mainland's ability to cope with natural disasters under complex meteorological conditions.
At present, "supercomputers" are widely used in molecular dynamics simulation in pharmaceutical research and development, fuel cell design in electric vehicles, aerodynamic shape optimization in aircraft design, boundary stability calculation of restraint devices in the field of nuclear fusion, and quantum mechanical technology in atomic physics.
(Image source: Veer Gallery)
Supercomputers also have limitations
However, the construction of "supercomputers" requires not only billions of yuan in input costs, but also hundreds of thousands of yuan in daily electricity bills. In addition, the huge volume occupying thousands of cubic meters and the complex water-cooled heat dissipation system also limit the further development of the "supercomputer". As a result, people began to rethink how to make computers achieve more data computing power in a smaller size and lower power consumption.
(Image source: Veer Gallery)
A natural idea is to integrate more transistors on the same chip, once one of the founders of Intel, Gordon Moore proposed "Moore's Law", that is, the number of transistors on integrated circuits doubles every 18~24 months, so that the data processing capacity of the CPU will double. That's why, even our thousand-dollar mobile CPUs now have more computing power than commercial computers a decade ago — because the density of transistors has increased.
However, "Moore's Law" can not be maintained forever, one of the reasons is that the lithography process for chip microcircuit processing has approached the optical diffraction limit of 2~3nm, and it has been difficult to further increase the transistor density under the condition of ensuring chip yield. In addition, as the size of the transistor continues to decrease, the leakage phenomenon between the electrodes and the huge heat dissipation problem will also lead to the performance failure of the chip.
So, how can we continue to improve the computing power of computers while reducing the size and power consumption of computers?
(Image source: Veer Gallery)
When computers encounter "quantum"
As the saying goes, when things are indecisive, quantum mechanics. If computers encounter quantum mechanics, can such imaginative "quantum computers" use the magic of quantum mechanics to handle complex problems that require exponential computing power?
We know that classical computers use binary operations, and each basic unit of computation can only be in a definite state of 0 or 1, and this basic computing unit is also called "bit". However, this also means that the density of transistors on the chip can only be increased to increase the number of "bits" of classical computers, thereby linearly increasing the computing power of the data. However, if we use "quantum computers", this vexing problem can be solved.
The basic computing unit of the "quantum computer" is called "qubit", which can be probabilistic in a state of 0 or 1 at the same time, that is, a "quantum computer" with N "qubits" can be in the possible state of 2 to the N power at the same time, and the N power of 2 will show exponential growth with the increase of N, so that it can have exponential powerful computing power.
Imagine if we had such a peculiar "qubit", then 1 "qubit" could act as 2 arithmetic units, 10 "qubits" could act as 1024 arithmetic units, and 100 "qubits" could act as about 1.27 to the power of 30... In this way, we can use a very small number of "quantum computers" with "qubits" to defeat the magic of computational problems that require exponential magnitude.
Schrödinger's cat – feel the charm of "quantum superposition"
Fortunately, this magic is provided by the "quantum superposition" in quantum mechanics, which as the name suggests, is the superposition of a quantum system that can be in multiple states at the same time before it is measured.
(Image source: Veer Gallery)
For example, let's assume that a cat is enclosed in an unobservable box, and there is a switch device in the box that triggers the release of highly toxic gases, and the switch is triggered by receiving a signal for the release of radioactive isotope decay. In this way, the decay of the radioisotope will trigger the gas switch to poison the kitten, while the kitten will survive without decay.
However, since radioisotope decay is probabilistic (assuming a 50% probability), this means that the life and death of the kitten is also a probabilistic superposition before the box is opened to observe. At this point, we found that there is no definitive way to describe the state of the kitten, because the occurrence of the two events "kitten alive" and "kitten dead" depends on whether the radioactive isotope decays. That is, there is a theoretical 50% chance that the kitten is still alive and 50% of the time that it is dead. Therefore, the kitten is in a superposition state of "kitten alive" and "kitten dead", and the probability of the existence of both states is 50%, which is the thought experiment of the famous "Schrödinger's cat".
Of course, once the box is opened for observation, the state of the kitten will be uniquely determined in "kitten alive" or "kitten dead", which indicates that this "quantum superposition" in quantum mechanics collapses to a definite state immediately after being observed.
The basic unit of operation of quantum computers - qubits
For classical computers, each "bit" representing the basic computing unit is achieved through the switch of a single transistor integrated on the chip, which can represent a 1 state when the transistor is on and a 0 state when the circuit is open. Similarly, "quantum computers" also need to find a suitable physical carrier as "qubits" to exert the magic of "quantum superposition" in the real world. The difference is that this physical carrier needs to maintain the superposition of 1 and 0 states during the calculation process, and how to find this magical "qubit" in the real world has always been the goal of scientists.
The effort paid off, and scientists finally found a physical carrier in nature that can maintain the superposition of both 1 and 0 states. This physical carrier is none other than an old friend we have known since middle school - charged ions.
Charged ions have two important properties, the first is that they have an electric charge, and we can capture any number of charged ions through the physical means of "electric field-magnetic field". Second, according to the theory of quantum mechanics, the energy of charged ions is discrete, that is, the energy inside the same charged ion has a sequential level, just like the orbit of a planet, this energy ordering method is called the energy level structure.
Scientific research has found that in this unique discrete energy level structure, specific two-energy levels can be selected to construct "qubits". Among them, the higher energy level can represent the 1 state, while the lower energy level can represent the 0 state, so that the probabilistic transition between the two energy levels can represent the superposition of 1 and 0. This kind of charged ion coding in nature is called "qubit", and this way of performing quantum computing is called "ion trap".
With the continuous advancement of technology, people have also begun to try to encode "qubits" with artificial physical systems. The study found that when the electronic circuit prepared by the lithography process is cooled to close to 0.015K, it can exhibit discrete energy levels similar to charged ions, and this artificial two-level physical system is also known as "superconducting qubits". The advantage of this "superconducting qubit" is that it is compatible with modern integrated circuit processes, so it has received a lot of attention from the industry.
However, it also requires extremely low temperatures below 0.015K – a little above absolute zero and colder than outer space, which requires a super "refrigerator" to provide extremely low temperatures. Moreover, each artificial "superconducting qubit" cannot be completely consistent, which puts forward higher requirements for the fineness of calibration and the accuracy of control.
In addition, physical systems such as neutral atoms, light quantums, quantum dots and more imaginative topological quanta have also been proposed as "qubits", and for now, "ion traps" and "superconducting qubits" are still considered by scientists as strong candidates for quantum computing.
(Image source: Veer Gallery)
Theory into reality - quantum computers are here!
Scientists estimate that at N≧50, quantum computers will have up to 2 to the 50th power, which would exceed the limits of all classical computers. That is, in dealing with certain computational problems, quantum computers will show "quantum superiority" or "quantum supremacy" over classical computers.
In 2019, a 53-qubit processor based on a "superconducting quantum computing" scheme suddenly appeared, which took only about 200 seconds to sample a specific random number, a computational problem that would take about 10,000 years to use the most powerful supercomputer at the time. This exponential level of computing power has not only brought about an increase in computing speed, but also brought a revolutionary impact on many traditional industries.
For example, the public-private key encryption (RSA) algorithm widely used in the modern financial industry is considered absolutely secure, because even the strongest supercomputers take about 80 years to crack the password, while quantum computers can brute-force it in only about 8 hours with their exponential computing power, which means that modern encryption systems based on traditional cryptography will face a huge impact from quantum computers.
(Image source: Veer Gallery)
Quantum simulation – "customized" for specific problems
It is worth noting that "quantum computers" only show efficient computing power for some quantum algorithms, and cannot replace classical computers to deal with daily office tasks. In addition, the quantum superposition of the "qubit" itself is also extremely susceptible to external interference and loss, and there is still a long way to go before quantum computers that can eventually achieve large-scale fault tolerance.
However, until a general-purpose "quantum computer" is finally realized, we can still build specialized machines to handle specific computing problems, such a dedicated quantum computer at this stage that we generally call a "quantum simulator", or simply a "quantum simulation".
In fact, specialized machines capable of handling specific calculations are not far away, and the large wind tunnels used in aircraft designs are an interesting example. Taking the aerodynamic shape optimization design in aircraft design as an example, computer simulation in the classical sense needs to mesh and discretize the aircraft and the nearby airflow, and calculate the force analysis and motion state of each mesh, and finally integrate all the computing grids to obtain the overall aerodynamic data of the aircraft.
(Image source: Veer Gallery)
In order to achieve a sufficiently small grid accuracy, the computing power of a "supercomputer" is often required to achieve short-term data operations, and further infinite discrete meshing analysis is impossible. In order to solve this calculation problem, wind tunnel simulation experiments are generally carried out directly in large wind tunnels using scaled down models of aircraft to visually verify the reliability of the aerodynamic shape of the aircraft.
In fact, the large wind tunnel itself at this time is a computer, as long as we enter different aerodynamic parameters can intuitively get the simulated force and motion state of the aircraft, but this "wind tunnel computer" is not the computer we remember, but it shows far beyond the ability of classical computers in the specific algorithm tasks designed by the aircraft.
(Image source: Veer Gallery)
This wonderful idea has also driven scientists to re-examine some of the most complex computational problems, such as molecular dynamics simulations of drug reaction processes, relativistic simulations of black hole collisions, and electron escape during nuclear fusion.
In fact, as early as 1982, physicist Richard Feynman proposed that "the computing resources required in quantum mechanics increase exponentially with the increase in the number of particles, and the best way is to use another, more controllable quantum system to simulate the calculation of an otherwise complex quantum system." ”
Simply put, for some computing problems that require exponential computing needs, we should no longer use the classical calculation method of 0 and 1 to solve, but should find another simple and controllable physical system to simulate the original complex problem equivalently, so as to avoid the huge waste of computing resources, which is the basic starting point of "quantum simulation".
(Image source: Veer Gallery)
(Image source: Veer Gallery)
Therefore, it can also be said that the "quantum computer" is also a generalized quantum simulation system, but the "quantum computer" uses the parallel computing characteristics of the quantum superposition state, and realizes the exponential data computing power through the "qubit" and a series of quantum logic gate operations. The "quantum simulator" belongs to the narrow quantum simulation system, which can complete the simulation of a specific complex quantum system in a simple and controllable way by constructing a physical model equivalent to the target system.
(Image source: Veer Gallery)
Therefore, it can be concluded that "supercomputer" is the resource optimization and integration of classical computers, "quantum computer" is the use of quantum mechanics in a new way of computing, belongs to a kind of future-facing general computing machine, and "quantum simulation" is also a new calculation method of quantum mechanics, but is a special computer machine that can simulate specific problems at this stage.
However, the emergence of both of the latter tells people an exciting truth - the quantum era is quietly coming, and it will profoundly change people's future production and lifestyle in an unprecedented way. This change is unimaginable, just as people in the 19th century used mechanical computing machines such as the abacus, and it was impossible to imagine that the computing power contained in electronic computers completely changed the way information was exchanged.
You might as well imagine how our world will change drastically in the future with the powerful exponential computing power of quantum computers?
Author: Luan Chunyang
Author's affiliation: Department of Physics, Tsinghua University
This article is produced by Popular Science China and produced by China Popular Science Expo, "Popular Science China" is an authoritative brand of science and technology in which the China Association for Science and Technology and all parties in the society use information technology to carry out scientific communication.