SFKP • Computer Encyclopedia 丨 Five changes in chip development
Posted May 26, 2020 • 15 min read
In November 2019, on a university campus in the suburbs of Hong Kong, China, a group of engineers are designing special computer chips that they hope can be used in the next generation of Chinese-made smartphones.
They designed optical communication chips, which use light instead of electrical signals to transmit information, which is needed for 5G mobile phones and other networked devices. One of the engineers was leaning on a chair wearing a Stanford T-shirt. He was the chief engineer and professor responsible for the project-Yu Jie, a "chip expert" at the Hong Kong University of Science and Technology.
"Politically, everything can be used as a bargaining power," said Yu Jie. "If these companies and countries start to have reservations about technology, then everyone will be hurt and there is no benefit from a technical point of view. ** Part of the funding for the Yu Jie project was provided by Huawei. This Chinese telecommunications and telecommunications giant has continuously become the center of an international political storm in the past year.
Many industry observers worry that this political storm may put the global technology supply chain at risk of collapse. In particular, China's chips(or semiconductors) are generally dependent on overseas companies and may face technological failures, which has also accelerated the self-research steps of the domestic chip industry chain.
What are the core technologies of the chip? Where is it now? What are the current difficulties of research and development in our country?
In this issue of "SFKP • Computer Encyclopedia" , let ’s take a look at "Core Things".
- The birth of "core"
Chips are a way of miniaturizing circuits in electronics, also known as integrated circuits or microcircuits, and are often manufactured on the surface of semiconductor wafers.
From 1949 to 1957, Werner Jacobi, Jeffrey Dummer, Sidney Darlington, and Yasuo Tarui were developed separately. The prototype of the integrated circuit, but the predecessor of the modern chip was invented by Jack Kilby in 1958, which included a bipolar transistor, three resistors and a capacitor, and thus won the 2000 Nobel Prize in Physics . **
The chip has two main advantages over traditional discrete transistors:cost and performance. **
The low cost is due to the fact that the chip prints all the components through photolithography technology as a unit, rather than making only one transistor at a time. The high performance is due to the small components and close proximity to each other, and the small size brings a short path, allowing low power logic circuits to be applied at fast switching speeds.
For this reason, only in the second half century of its development, chips have become ubiquitous. Computers, mobile phones and other digital appliances have become an indispensable part of the structure of modern society. The computing, communication, manufacturing and transportation systems of modern society include The Internet depends on the existence of chips.
Many scholars even believe that the digital revolution brought by integrated circuits or chips is the most important event in human history. The maturity of chip technology will bring a great leap forward in technology, whether in design technology, Or a breakthrough in semiconductor technology, both are closely related.
- Five changes in chip development
Although the prototype of integrated circuits appeared as early as 1949, until the emergence of bipolar and MOS circuits in the 1960s, With the development of silicon planar technology, the form of electronic products manufactured by tubes and transistors has undergone qualitative changes from quantitative changes. Chips have truly become a relatively complete industry.
In the long development period of more than 60 years, the chip industry has undergone four major changes. The important reasons for these four changes include "In order to solve the problems in system design or chip circuit design", also let the chip There have been structural changes in the industrial division of labor.
The first change:the industry's first shuffle
The first revolution in the integrated circuit industry began in the 1970s. With the birth of microprocessors and memory, the era of vertical integration in which system companies monopolized the design of systems and ICs was transformed into a system of division between system companies and IC companies.
From 1960 to 1970, the industry shuffled for the first time.
In this decade, basically every chip manufacturer needs to carry out all chip-related design and manufacturing. However, with the development of chips, more and more functions, related processes and materials are involved, and the design process takes longer and longer. The chips of many manufacturers with lower productivity have become "expired products" before the official launch ".
In 1970, many components of the chip began to be standardized gradually, which gave many small and medium-sized manufacturers the opportunity to focus on chip integration and manufacturing. Through the direct use of standardized components, the speed of chip production iteration was accelerated.
At this stage, vertically integrated manufacturing companies(IDM) companies play a major role in the chip market, while integrated circuit design(IC) exists as a subsidiary.
However, this also caused a problem that IDM companies attached too much importance to the control of R & D costs, thereby sacrificing the ability to innovate in the chip manufacturing process.
The second change:product standardization, clear division of labor in the industry
The second change was in the 1980s. In order to improve the speed of the system and simplify the program, the application-specific integrated circuit(ASIC) technology came into being. Engineers can directly design gate arrays(CPLD), programmable logic devices(FPGAs) using logic gates without understanding the details of transistor circuit design. Standard units, fully customized circuits, etc., which further standardize the design and production of chips.
However, there are still many independent IC circuit designs on the entire chip circuit, which is difficult to meet the system cost, reliability and other requirements of the entire customer. At the same time, customers require continuous increase in IC integration, reduction in chip area, cost reduction and efficiency enhancement, thereby enhancing the competitiveness of products, and gaining more market share and richer profits.
Changes in design concepts have led to the emergence of professionally-designed Fabless companies, and the foundry + Fabless operation model is filled by the professional wafer foundry Foundry to fill the capacity required by Fabless companies.
The third change:the pace of the market is accelerating and the industry's capabilities are being upgraded
The third change was in the middle and late 1990s. With the continuous advancement of Fabless, the number of transistors integrated on the chip has reached tens of millions, and the process has also been developed to 180nm, which has made the chip's capabilities and application scenarios extremely high. Big enhancement.
At the same time, the emergence of single chip system(SoC) has significantly improved the design capabilities of ASICs. The SoC design methodology came into being, and it contains three contents:
1 . System design method
2 . Design and use of IP core
3 . Deep submicron integrated circuit design
With the widespread adoption of SoC design methodologies, chip design companies purchase the IP of third-party companies and combine them into SoCs. The entire process is like building blocks. The scale of the chip has grown exponentially, from millions to today. One billion gate level.
Moreover, highly complex system functions and increasingly fast time-to-market requirements do not allow chip designer companies to start from scratch. They must learn from and use mature designs for their product development services. This also determines that the SoC design must adopt a different method from the traditional monolithic integrated circuit design.
At this stage of the development of the integrated circuit industry, professional division of labor has initially taken shape. With the gradual maturity of SIP design, EDA tools, chip design, wafer manufacturing, packaging, and testing, a large number of excellent companies have emerged in various professional links.
The fourth change:the integration of the chip industry
With the maturity of the chip industry, it has now entered the development stage of clustered virtual vertical integration(CVVI) mode. Cluster, virtual vertical and integration are the essence of this stage.
Specifically, it is to allow companies specializing in different fields to complement each other through alliances or strategic partnerships, so as to achieve the strategic purpose of rapid layout and then achieve effective vertical integration. **
In the field of chips, in order to shorten the chip design cycle, companies in the industrial chain must communicate and cooperate with each other. Through the links between manufacturers and IP integration, keep up with customer needs and market changes, so that each other's benefits can be maximized The best competitive advantage.
As "heterogeneous integration" and "Chiplet" become the future development trend of the integrated circuit industry, new business models will emerge. Now, the single-player model has become weak, and clustering, virtual vertical, and integration have become the general trend.
Fifth change:the next decade-AI chip development
In April 2017, Google announced a paper to be published in ISCA2017:"In-Datacenter Performance Analysis of a Tensor Processing Unit", which disclosed the details of its AI chip; May 2017, GTC2017 conference, Nvidia CEO Huang Renxun also released the GPU of Volta architecture.
Because of these two important announcements, 2017 was defined as the first year of the AI chip.
The development of chip technology is a necessary condition for the development of artificial intelligence. Chips represent computing power. There are several important pillars for the development of artificial intelligence, including
-Data:facts or observations
-Algorithm:method to solve the problem, such as deep learning algorithm
-Computing power:computing power
In terms of data, big data accumulates at a high speed in the Internet era, and the amount of data that needs to run increases sharply, but the existing computing power is difficult to match; from the algorithm side, even if there are logically available algorithms, it needs sufficient computing power to process data samples 3. Train the machine. Due to the indispensability of computing power and the slowdown in the growth of communication applications(smartphones, etc.), the development of artificial intelligence will also become the main driving force for the progress of chip technology.
On the whole, if divided according to the design concept, AI chips can be roughly divided into two categories.
The first category is "AI acceleration chips", which deterministically accelerate certain types of algorithms or tasks, so as to meet the requirements of the target application field in terms of speed, power consumption, memory footprint, and deployment costs. At present, there are two main ways to develop AI accelerated chips:one is to use existing GPU, many-core processors, DSP, FPGA chips for software and hardware optimization; the other is to design a dedicated chip, that is, ASIC .
The second category is "smart chips", which allow chips to learn and derive using different AI algorithms like humans, handle a series of tasks including perception, understanding, analysis, decision-making, and actions, and have the ability to adapt to scene changes.
At present, there are two types of design methods for the research of smart chips for comprehensive and adaptive capabilities. One is a "neural mimic chip" based on brain-like computing; the other is a "software-defined chip" based on reconfigurable computing. **
Around these two directions, the world's major chip companies are actively deploying in the field of artificial intelligence. It can be said that the time to greatly enhance the development of artificial intelligence through chip technology is very mature.
However, judging from the general trend of chip development, is still in the initial stage of the development of AI chips, whether it is scientific research or industrial applications, there is huge room for innovation.
It is an inevitable direction for scientific research to develop from AI acceleration chips that determine algorithms and fields to smart chips with higher flexibility and adaptability. Industry experts also issued a judgment that the next decade will be an important period for the development of AI chips, and it is expected to make a huge breakthrough in architecture and design concepts again.
Third, the chip production process and materials
The instability of the international situation in the past two years has caused many people to worry about China's chip production.
Due to the lack of key technologies and equipment, many products in China, including mobile phones, switches, network communication base stations, and core chips, cannot be produced independently. This is also one of the reasons why the United States relies on "yaowuyangwei".
So, what core technologies are involved in making a chip? What raw materials are needed?
How is the chip made?
The complete process of chip production includes chip design, wafer production, packaging production, testing and other links, of which the wafer production process is particularly complicated.
1 . Manufacturing silicon wafers
Add carbon to the sand and convert it into silicon with a purity of about 99.9%under the action of high temperature. After melting, a pencil-like silicon pillar is pulled out of it. The silicon wafer is cut into round skins with a diamond knife, and then polished to form a silicon wafer. The common diameters of silicon wafers are 8 inches and 12 inches. The larger the diameter and the thinner the wafer, the lower the cost of a single chip. However, the processing difficulty and process requirements are higher.
2 . Lithography
The basic process of the photolithography process is to first coat a layer of photoresist on the surface of the wafer(or substrate) and dry it. The dried wafer is transferred to the lithography machine.
Apply photoresist on the silicon wafer, let the ultraviolet ray irradiate the photoresist through a mask, and the mask is printed with pre-designed circuit patterns. During the lithography process, the photoresist exposed to ultraviolet light is dissolved by photochemical reaction. Some chip manufacturing processes require a second baking of the exposed wafer, which is called post-exposure baking. Post-baking is The photochemical reaction is more complete.
Finally, the developer is sprayed onto the photoresist on the surface of the wafer to develop the exposure pattern. After development, the pattern on the mask is left on the photoresist. Gluing, baking, and developing are all done in a uniform development machine, and exposure is done in a lithography machine.
The entire exposure and development system is closed, and the wafer is not directly exposed to the surrounding environment, so as to reduce the influence of harmful components in the environment on the photoresist and photochemical reactions.
3 . Doping
The ion implantation imparts the characteristics of the silicon transistor. The specific process is to start from the exposed area on the silicon wafer and put it into the chemical ion mixture.
Generally, boron or phosphorus is implanted into the silicon structure, and then copper is filled to interconnect with other transistors. Then you can apply another layer of glue on top to make another layer of structure. A simple chip can use only one layer, but a complex chip usually has many layers. Most of the current chips contain dozens of layers, forming a three-dimensional structure.
4 . Package test
After the chip is ready, the chip is cut from the wafer with a fine cutter, soldered to the substrate, the manufactured wafer is fixed, the pins are bound, and various packaging forms are made according to the needs. That is why the same kind of chip core can have different packaging forms. After testing, it can be packaged and sold.
What are the core materials of the chip?
1 . Mask
We know that one step in the chip manufacturing process is "lithography", which mentions the concept of a "mask", that is, a photomask.
The photomask is made according to the pre-designed circuit diagram. After the light source of the lithography machine passes through the photomask, the circuit diagram is printed on the wafer, and the photomask is like a negative.
In 2018, the global semiconductor manufacturing materials market was US $32.2 billion, of which the mask market was approximately US $4.5 billion, accounting for 14%of manufacturing materials, second only to silicon wafers.
2 . Photoresist
In addition to the mask, we also mentioned photoresist. To complete the photolithography, a photoresist must be coated on the silicon wafer in advance.
Photoresist, also known as photoresist, refers to an etch-resistant thin film material whose solubility changes through irradiation or radiation of ultraviolet light, electron beam, ion beam, etc. In terms of nature, photoresists are divided into positive photoresists and negative photoresists. It is the positive photoresist that increases the solubility when exposed to light, and the negative photoresist decreases the solubility.
Regardless of whether it is negative, in short, it is necessary to make the areas that are exposed and not exposed to form areas of different solubility, and then dissolve the unwanted areas.
Different types of chips and chips with different manufacturing processes use different photoresists and different levels of difficulty.
By 2018, the global photoresist market is $1.6 billion, which is not large, but it is crucial.
In addition, the photoresist market is also a highly monopolized market, with the market share of the top 5 companies in the world exceeding 85%. The top four photoresist companies are Japanese companies, namely synthetic rubber, Shin-Etsu Chemical, Tokyo Yinghua and Sumitomo Chemical.
3 . Electronic gas
The circuit is engraved on the silicon chip. To make the chip work, there is a premise that the transistors engraved on the silicon chip must have switching characteristics. For a circuit to have switching characteristics, ion implantation is required, and ion implantation requires electron gas, also known as electron special gas.
The first characteristic of the electronic gas used on the chip is high purity, and the purity of most electronic gases is more than 99.99%. And most electronic gases have the characteristics of high pressure, flammability, high corrosion and high toxicity.
Electronic gas is also a highly monopolized market. Major companies include American Air Chemicals, Praxair, Germany Linde Group, France Air Liquide, and Japan Dayang Risui Co., Ltd., which account for more than 90%of the market.
4 . High purity sputtering target
When engraving the circuit diagram on the wafer, doing various cleanings, and implanting ions, but it is necessary to connect the electronic components on the chip, just like you put all the components in place, and then you have to use Wires connect them to each other, which requires sputtering. Sputtering is mainly to prepare thin film materials, which is a kind of physical vapor deposition(PVD) technology.
The general meaning of sputtering is to bombard the target with ions, then the atoms on the target are bombarded, and finally fall on the single crystal silicon substrate, and then form a metal layer with a specific function, thereby forming a conductive layer or a barrier layer This is metallization. After the metal layer is formed on the chip surface, then lithography or etching is used to remove the unnecessary parts, so the thin metal lines are left on the chip surface, which can connect various components on the chip.
High-purity sputtering targets include aluminum targets, titanium targets, tantalum targets, and tungsten-titanium targets. Among them, aluminum targets and titanium targets are mainly used in 8-inch wafer production, and tantalum targets and copper targets are mainly used in 12-inch wafers.
Sputtering targets also belong to the oligopolistic market. Major companies include Japan Nippon Metals, Honeywell, Tosoh, Praxair and Sumitomo Chemical.
Overall, the barriers to chip manufacturing are very high, and the related materials are mostly in the monopoly of the industry, so the overall barriers of the industry are high. With the influence of the political situation, it is indeed easy to become a resource of restraint among various countries.
Postscript:core things are hard to predict
The substitution of localization has become a banner leading the development of the core technology industry in China for more than 40 years of reform and opening up, and it is also a revolution. In the corner of the blueprint for this revolution, the domestic marathon of the semiconductor industry has been sprinting for many years, from upstream materials and equipment to midstream design and manufacturing, and then downstream packaging and testing, the domestic development and competition of all links in China's semiconductor industry chain are also extremely fierce.
"There is no doubt that China has engineers who make chips. The question is whether they can produce competitive products." Spirit Valley historian and artificial intelligence scholar Piero Scaruffi(Piero Scaruffi) Asked.
But industry analysts like Scalufi still have doubts about China's true innovation capabilities. He believes that China's current technological success lies in technology application, not technology creation.
"If your measure is how many people use smartphones to shop, then China wins. But if your measure is the Nobel Prize winner, then China loses terribly," he said, "Of course, China is very successful in applying technology to greatly change society. "
The US regulation on the one hand does have a great impact on China's chip industry, on the other hand it also gives Chinese companies an opportunity to prove their strength.
For example, Huawei HiSilicon and Ziguang Group have broken through technical limitations in chip design, especially Huawei HiSilicon has mastered 7nm chip design technology. For chip manufacturing, SMIC's 14nm process technology has been officially put into production. In the field of lithography machines, Shanghai Microelectronics has taken the lead, while in the field of etching machines, China Microelectronics and North Huachuang have both attacked. The 5nm etching machine independently developed by China Microelectronics has also been applied to TSMC's first 5nm chip production line.
In addition, China's chip packaging and testing industry is "a single company". At present, the world's largest packaging and testing company is ASE Group of Taiwan, and the world's third packaging and testing company is China's Changdian Technology(600584). Its market share is similar to the first.
But as Scalufi said, compared to chip manufacturing, the technical barriers to packaging and testing are lower, so entry is relatively easy. On the other hand, packaging and testing is a labor-intensive industry in the field of chip segmentation, and in terms of labor, China has always had an advantage.
So, the two largest economies in the world are colliding again, we cannot be blindly optimistic, nor do we need to be pretentious.
With the reshuffle of the world's political landscape, who knows whether the chip industry will usher in a new reshuffle.
CCTV News:"How are chips made? 》
Core Idea:"Six Era of Chip Development"
Junlin Research Center:"How important is the domestic substitution of chip manufacturing materials? 》
Southern Weekend:"The Man on the Chip"