Chips: The power to change the world and the challenges of the future
As the core of modern scientific and technological development, the importance of chips is self-evident. As the famous science and technology writer Wang Bo said in his award-winning work "A Brief History of Chips", the invention and development of chips have been driving the progress of human society. From the advent of block integrated circuits in the 60s of the 20th century to today's ubiquitous use of chips in various fields, the rapid development of chip technology has profoundly changed our way of life.
The advent of chips first revolutionized the face of computers. Early tube computers were bulky, power-consuming, and had limited computing power. The advent of chips has enabled computers to become miniaturized, consume less power, and increase computing power by orders of magnitude. The popularity of everyday electronic products such as personal computers and smartphones depends on breakthroughs in chip technology. Without chips, there would be no information age today.
In addition to the computer field, the development of chip technology has also promoted the transformation of many industries such as communications, medical care, and transportation. High-speed broadband networks, 5G communications, medical imaging diagnostic equipment, etc., are inseparable from the support of chips. The realization of emerging technologies such as autonomous vehicles and drones also depends on the powerful computing power of chips. The development of cutting-edge technologies such as artificial intelligence and quantum computing in the future will also be highly dependent on breakthroughs in chip technology.
At a time when chip technology has brought us earth-shaking changes, its own development is also facing some serious challenges. As Wang Bo pointed out in his speech, the future development of chips faces the obstacles of "three walls" of "memory wall", "power consumption wall" and "tunneling effect".
Memory wall: The storage vs. compute divide
The so-called "memory wall" refers to the fact that the CPU accesses the memory far behind its own computing speed. This is due to the existence of a "memory wall" between the CPU and memory, which makes the transfer of data between the two extremely inefficient.
The root of this problem is that CPU and memory are two completely different hardware architectures. A CPU is a high-speed logical computing unit, while memory is an array used to store data. The CPU needs to constantly read data from memory, perform calculations, and then write the results back to memory. But because of the huge difference in how they work, transferring data between the two is as difficult as having to go through a high wall.
For example, when a CPU needs to read data from memory, it must first put the memory address into a dedicated register. The memory controller then reads the address and transfers the data to another buffer. Finally, the CPU can read the required data from the buffer. This process involves multiple copies of data, which is extremely inefficient.
The "memory wall" problem limits the performance of the chip to a certain extent. Because no matter how powerful the CPU is, if it can't read and write data from memory efficiently, the overall system performance won't be fully utilized.
Power walls: The challenge of energy consumption is growing
Blocking the "power wall" is also a major challenge for chip development. The so-called power wall refers to the increase in the number of chip transistors and the shrinking of the process, the power consumption of the chip is also increasing, which brings huge pressure on heat dissipation and power supply.
Excessive power consumption will not only lead to problems such as excessive heat generation and reduced battery life, but will directly limit the room for improvement of chip performance. Because the power consumption is too high, the chip operating temperature will be too high, so the chip has to reduce the operating frequency and voltage to avoid damage.
For example, NVIDIA's latest RTX 4090 GPU chip consumes up to 450W, which is equivalent to the overall power consumption level of a medium desktop. This large power consumption is a challenge for both the cooling and power supply systems. In the future, if the power consumption of chips continues to increase, it is likely to hit a physical ceiling.
To solve the power wall problem, it is necessary to start from chip design, manufacturing process, heat dissipation system, power management and other aspects. For example, the adoption of new transistor structures such as FinFETs, the introduction of new low-power interconnect technologies, and the optimization of the functional partitioning of chips are expected to help reduce the power consumption level of chips.
The tunneling effect: the challenge of quantum effects
The final "tunneling effect wall" comes from the challenge of quantum physics. The so-called tunneling effect means that when the size of the transistor is reduced to the atomic level, there is a certain probability that the electrons will "tunnel" out of the potential barrier of the transistor, causing the transistor to fail.
This is because, when the size of the transistor reaches the order of nanometers or even smaller, quantum effects become very noticeable in it. According to quantum mechanics, an electron is not only a particle, but also has the properties of a wave. When an electron encounters an energy barrier, although its kinetic energy is not enough to cross the energy barrier, due to the nature of the wave, it still has a certain probability of being able to "tunnel" through the energy barrier.
In the case of transistors, this tunneling effect means that electrons escape, making the transistor unable to function properly. The tunneling effect sets a physical limit to the further shrinkage of the chip.
One possible way to solve the tunneling effect is to explore new transistor structures, such as the use of new materials and device structures such as graphene, nanowires, or quantum dots, to avoid the effects of quantum effects. On the other hand, some researchers have proposed a new computing architecture of "transistor without chips", which is completely free from the limitations of traditional transistors.
Breakthrough innovation: the driving force of scientific and technological progress
In the face of the challenge of the "three walls", the development path of chip technology is not dark. Whenever the development of science and technology encounters a bottleneck, innovation will emerge and open up new breakthroughs.
In the case of memory walls, a variety of new storage technologies have emerged to alleviate the data transfer bottleneck between the CPU and memory. For example, high-bandwidth memory (HBM, 3D chip stacking, Storage Class Memory (SCM, etc.) have improved the transfer efficiency between storage and computing to varying degrees.
The rise of heterogeneous computing architectures also provides a new way of solving the problem. In a heterogeneous architecture, different computing tasks are assigned to the most suitable processor to execute, so as to give full play to the advantages of each processor and avoid the inefficient transfer of data between the CPU and memory.
For the challenge of power wall, in addition to some of the above-mentioned technical means to reduce power consumption, the development of quantum computing has also brought us a new dawn. Compared with classical computing, quantum computing has the advantage of low power consumption, and the emergence of quantum computers in the future will greatly alleviate the pressure on the power consumption wall.
As for the tunneling effect, while it sets a physical limit for further shrinkage of chips, it does not mean that Moore's Law ends. On the basis of the existing process, we can continue to improve the performance and integration of chips through heterogeneous integration and three-dimensional integration.
Technological developments are often unexpected. Just as the invention of the transistor revolutionized the vacuum tube computer back then, a new computing architecture may emerge in the future that will revolutionize the current chip design paradigm.
As Wang Bo said, "The best way to predict the future is to invent it." "The "three walls" faced by chip technology are of course severe, but as long as we maintain the courage and motivation to innovate, we will be able to break through many obstacles, continue to promote the progress of science and technology, and change the world.
The development of chip technology has been promoting the progress of human society, from computers, communications to transportation, medical and many other fields, have been or are being changed by chips. However, the development of chip technology itself also faces three major challenges: "memory wall", "power consumption wall" and "tunneling effect", which limit the further improvement of chip performance to a certain extent.
The history of technological development tells us that challenges are not a desperate situation, but a driving force for innovation. Whether it is new storage technology, low-power design, quantum computing, or new computing architecture, it has opened up new possibilities for us to break through the "three walls". As long as we maintain the courage and motivation to innovate, chip technology will be able to continue to develop and bring more changes to mankind.