We can say that the Moore’s Law (transistor density doubles every two years) ended in April 2016 when Intel failed to produce 10nm devices. In March of the same year, the death of Dr. Grove, who laid an important foundation for the determination of Moore’s Law, ended this epic era in human history. Moore’s Law is important because it has guided and coordinated the development of the integrated circuit industry for 40 years. At present, the overall annual revenue of the integrated circuit industry is between 400 billion and 500 billion US dollars, and the entire electronic information technology industry is between 40,000 and 5 trillion US dollars. It can be said that this industry born in the middle of the last century has become the cornerstone of the development of human civilization.
The development of integrated circuits requires adjustments and changes
When the Moore’s Law era will end, the industry has always had different predictions, and at the same time, it is also making corresponding plans. Dr. Gordon Moore himself has been very cautious. In the past 30 years, every time he was asked when Moore’s Law would end, he kept answering: “There are still 10 years.” Before 2010, the main bet against Moore’s Law was invalidated. Although most experts cannot see specific solutions or even the main development direction, they all have confidence in the next generation of engineering research and development and are happy to see themselves surpassed.
Major innovations in history that mainly rescued Moore’s Law include complementary metal oxide semiconductors to replace bipolar transistor devices to reduce power consumption; ultra-ultraviolet 13nm to replace 193nm step projection exposure, step projection to replace mask alignment exposure, and to enhance pattern fineness Ion implantation replaces diffusion to improve doping accuracy; high dielectric constant materials replace SiO2 to enhance device speed and reduce leakage; low dielectric constant materials replace SiO2 and copper metal processes replace aluminum to increase interconnection line speed and reduce power consumption; Chemical mechanical polishing increases the number and complexity of interconnection lines, organic packaging enhances winding density and reduces costs; FINFET enhances switching speed, and so on.
The integrated circuit industry currently has the following pattern: manufacturing centers from the United States to Japan to now China Taiwan and South Korea, the materials center is mainly in Japan, the entire manufacturing chain is in the hands of the United States, and low-end auxiliary manufacturing is in mainland China. The core issues of integrated circuits in history include computing power, power consumption, cost, reliability, manufacturability, etc. These issues will continue to run through the development of integrated circuits in the post-Moore’s Law era in the future.
The era of Moore’s Law will eventually pass, because exponential growth cannot last forever. At the rate of doubling the transistor density every two years, the number of transistors in a single processor will exceed the number of human brain cells in 30 years, the total number of atoms in the earth will exceed the total number of atoms in 300 years. More realistic are the two physical limits that are about to be faced: the Landauer limit (0.02eV) and the quantum tunneling limit (2nm). The former means the energy consumption per unit operation is close to the limit of thermodynamic perturbation, and the latter is in the probability of penetrating the barrier is greatly increased at the size, causing the transistor to fail. These two limits are reached in almost 20 years according to the linear extrapolation of Moore’s Law. Therefore, the current industrial development requires extremely significant adjustments and changes. IBM released the so-called 2nm device technology in 2021 (the actual mass production may be 6 years), and its source and drain spacing is actually 12nm. Ingenious engineering innovation and naming games cannot avoid the actual physical limits.
In addition, there are many ideas that failed at that time, and they are finally gradually becoming mainstream solutions in the post-Moore’s Law era. For example, in the 1980s, the Japanese government funded 3D transistor stacks. AMD used more because of poor mass production yields. A physically cut silicon chip to improve the performance of the chiplets technology, Intel’s fan-out packaging process for the 20GHz processor (because the heat dissipation problem cannot be solved), etc. These projects and later innovations (such as quantum computing, non-binary, non-Von Neumann architecture, artificial intelligence, and machine learning) were shelved due to the computing power requirements, process capabilities, industrial chain maturity, market acceptance, etc. at the time. ), which constitutes the solution framework for the post-Moore’s Law era.
Today, the industrial center is gradually transitioning from a hardware center to a software center, and from a general-purpose computing center to a dedicated computing center. The innovation of the hardware industry has more influential factors, which are no longer relatively independent of applications, software, algorithms, architecture, devices, and packaging, but are intertwined with each other and require comprehensive consideration.
