The “Four Levels” of Chip Design Companies

Latest update time：2021-03-07

Reads：

Source: Content Reprinted from the public account [ Wai Rui Lao Ge ], author Wai Rui Lao Ge, thank you.

In the martial arts world, the realm of swordsmanship is divided into four levels;

First level: "There is a sword in the heart, but no sword in the hand"

This is the state of a beginner. One has sword techniques in mind, but they are not yet fully developed and cannot be used at will.

The second level: "There is a sword in the heart and a sword in the hand"

The sword techniques are pure numbers, like the movement of fingers, which can be freely controlled, and the eyes and hands can move freely, and the world can be conquered;

The third level: "There is no sword in the heart, but there is a sword in the hand"

The swordsmanship is internalized in the heart and externalized in action. It is not limited to the swordsmanship moves. No moves are better than moves.

The fourth level: "There is no sword in the heart, and no sword in the hand"

Not being bound by objects, grass, wood, bamboo, and stone can all be used as swords, reaching the state of unity between man and sword;

This is how sword skills are advanced, and the same is true for chip design companies;

The first level: "There is a core in the heart, but no core in the hand"

Boss Jia is a wealthy man. Seeing that the chip industry is so hot, Boss Jia, who does not want to miss any opportunity, decided to enter this industry. Boss Jia is not only wealthy, but also has many connections. Boss Jia's connections are unfathomable. According to Boss Jia, he has an AI chip project that needs to identify criminals from the images in the camera and then transmit them to the backend for alarm. Boss Jia's company used to buy chips from other manufacturers to do this, but it was expensive and hard to get them. Boss Jia can sell tens of millions of chips every year through various channels. Boss Jia calculated that it is more cost-effective to make chips by himself. So he recruited a group of people and established a chip company.

Boss Jia is currently in a state of "having a chip in mind but not in hand". At this level of companies, the main focus is on the process of turning PPT into chips; this entry seems simple, but it is actually very complicated. Boss Jia wants to develop this AI chip, the core application of which is the image collected by MIPI, cached in DDR, identified as a criminal through the AI processor, and then the image is controlled by the CPU to upload to the cloud through the WIFI network. When the team started working, he found that he faced two major problems:

The first question: IP selection: There are IP vendors for CPU, DDR, bus, AI processor, MIPI, WIFI in SOC, but which one to choose is a science? In addition to price, are there any successful application cases for these IPs? Can the speed meet the requirements? Especially for AI processors, should we choose the IP of the vendor or develop it by our own team? How to choose?

The second question is the choice of process: The choice of process is based on the expected number of chips, the one-time NRE cost, whether the process can meet the performance requirements, whether there are options for analog IP under the process, and the power consumption. This is a comprehensive equation to be solved;

Boss Jia's company solved these two problems; then began the design phase. The team Boss Jia hired mainly worked on SOC integration and verification. This type of chip company does not have high requirements for design, but has great demand for verification, the so-called "not strong in design, but extremely high in verification"; the back-end can rely on back-end design services; thus, it can be delivered to the market faster; the chip team mainly completes the RTL-to-netlist part, and the rest is delivered to the back-end design service company to complete the netlist-to-gds part. The back-end design service company will find TSMC/SMIC for wafer fabrication, and then find Changdian/ASE for packaging and testing; as expected, Boss Jia got the chip three months after the design was completed;

It usually takes an average of 18 months for Boss Jia's chip company to cultivate this level, and a talented person usually needs 12 months;

If you finally get the chip after going through 81 difficulties, and it can bring up (the chip lights up), it means that you have made some progress in this level of mental training and can proceed to the next level of training;

If after going through all the difficulties, the chip you get is a stone, you will need to start over from the beginning and cultivate this layer again until you succeed;

This level of cultivation is successful. Depending on the size of the chip, the difficulty and time of cultivation are different. Generally, it is normal to have gone through several tape-outs.

The second level: "There is a core in the heart and a core in the hands"

If you were asked to choose between Duan Yu who knows Six Meridians Divine Sword and Qiao Feng who knows Eighteen Dragon Subduing Palms to duel Ding Chunqiu in a game, a normal person would definitely choose Qiao Feng, because Qiao Feng is a sure win, while choosing Duan Yu is risky. Duan Yu's Six Meridians Divine Sword is invincible and extremely exquisite, but its output is extremely unstable, sometimes working and sometimes not, which can easily cause people to collapse at critical moments.

If you develop a chip, you also hope that the final result will be Qiao Feng's "Eighteen Palms of the Dragon Subduing" instead of Duan Yu's "Six Meridians Divine Sword". After 18 months, if you find that the chip can work like "Schrödinger's cat", it may work or not work at the moment of power-on. What will be Boss Jia's reaction at this time? Boss Jia will collapse. In order for Boss Jia not to collapse, there must be a way to allow chip design companies to output chip products stably and reliably.

Chip design itself is a standard industrialized process with a complete set of process controls. This set of process controls can ensure two points:

1: What you think is what you get: The chip can be developed and designed smoothly according to the product definition, and it will definitely work;

2: The result is qualified; the chip is qualified after passing the test (DFT, BIST, functional pattern), which guarantees that the chip can be 100% qualified.

After completing this stage and achieving a relatively good yield rate, such as 95%, it can be considered as completing the step from entry to proficiency;

What you think is what you get, which requires R&D process control, such as IPD, a complete set of R&D management processes, including input from market research, product definition input, architecture design, IP selection, coding specifications, verification specifications, synthesis and back-end signoff processes; these specifications are the lessons learned from countless tape-out bugs and failures. Complying with these processes will greatly reduce risks, otherwise you will fall into one pit after another; there are countless chips that fail due to product definition deviations, insufficient design redundancy, insufficient verification, and non-standard back-end signoff specifications;

Currently, 90% of China's chip companies have less than 100 people, and even fewer R&D personnel, and they basically adopt a guerrilla R&D approach; they mainly rely on the project manager's own capabilities, and the project manager's focus is generally on the quality of RTL, the quality of verification (functional coverage, code coverage, etc.), and the quality of the netlist. Back-end process control basically relies on the back-end company's own processes. If the back-end service company is more standardized, the chances of success are greater. If the back-end service company's own processes are not standardized and the signoff process is not perfect, then the risk is very high. From relying on people to relying on processes is a big change;

The first stage is the first stage. The management of packaging and testing, production capacity, inventory, and the entire supply chain is more arduous and tests the accumulation of chip companies in this area. Chip design is a high-end manufacturing industry, and none of the problems of the manufacturing industry can be avoided. If the supply chain is not done well, a phenomenon will occur. If the market is booming and the production capacity is not estimated and coordinated well, the opportunity to make money will be missed. If there is too much inventory and the market does not pick up, it will lead to a backlog of costs.

After multiple tape-outs at the second level, the chip design company can continuously fine-tune the process and implement it throughout the entire R&D process, so that it can truly achieve the goal of hitting wherever it is pointed, and the chip team can have combat effectiveness output; it can be called a reliable chip team, rather than a gang or guerrillas;

This level of cultivation depends on the team's understanding. It may take 1-2 years or several years to complete. It may even never be completed because the money for tape-out is spent and the team is disbanded. There are many examples. For rich people like Boss Jia, who have sufficient capital investment and experience in recruiting talented people, it is also possible to succeed in cultivation. By reaching this level, chip companies can delve deeper into their main channel, accumulate core technologies, achieve higher performance, lower power consumption, and less cost, including technologies that others cannot handle, and gradually win the market and gain growth opportunities.

The third level: "No core in the heart, core in the hand"

Chip companies are essentially businesses. Providing customers with stable, reliable, and price-competitive chips and obtaining funds from customers is normal business logic. However, taking more advanced means to acquire customers is the third level of cultivation.

"No chip in mind, but chip in hand", this realm is rarely achieved by domestic companies. Manufacturers at this level have gone beyond the level of selling chips and directly provide solutions, patents, and ecology, referred to as SPE;

The solution level is relatively easy to learn. For example, MediaTek in Taiwan, China, also provided a complete turnkey solution in the era of copycat phones. Small factories in Huaqiangbei did not have to worry about how to design circuits. They just added a screen and battery to the solution provided by MediaTek, and used MediaTek's mobile phone chips to make a complete mobile phone. This solution lowered the threshold for mobile phone design, laid the foundation for the prosperity of copycat phones, and also made MediaTek chips have the reputation of "the king of copycat phones";

The same is true for HiSilicon. Before HiSilicon, there were many system software manufacturers that provided license plate recognition, and they were all doing well. HiSilicon made a smart parking system based on its own IPC SOC and opened it for free in the form of SDK. Soon many downstream manufacturers followed suit, which lowered the threshold of the entire industry. Therefore, the chip defeated a series of competitors such as TI and won market recognition. This kind of free solution made its own solution the mainstream in the industry on the one hand, and the downstream manufacturers lowered the threshold and made profits on the other hand, so the chip was a natural thing.

Patents are one level higher than this. Qualcomm has deployed a large number of patents in the 3G/4G/5G fields, and can levy Qualcomm tax on each chip. Qualcomm has 90% of patents in 2G CDMA technology, about 27% of patents in 3G WCDMA technology, about 16% of patents in 4G LTE, and about 8% of patents in 5G. A smartphone will definitely involve patents in these technologies, so Qualcomm can charge all mobile phone manufacturers. In addition to the chip money, mobile phone manufacturers also have to pay patent fees for their Snapdragon processors. It's like skinning a cow twice. Qualcomm is like Ding Chunqiu's "Chemical Power Method", making a lot of money through patent taxes.

If Qualcomm's taxation is unpleasant, then achieving market leadership through ecosystem building is a smarter way.

The first successful example is Intel. Intel is a platinum member of the Linux Foundation, the most famous open source software, and the OpenStack Foundation, the most successful cloud computing open source project. It is also an important member of open source organizations such as the Apache Foundation, the GNOME Foundation, the Eclipse Foundation, the Open Data Center Alliance, the Open Fabric Alliance, and the Open Virtualization Alliance. It is also one of the largest contributors to many open source projects such as Linux, ChromiumOS, and OpenStack. Some people say that Intel is a software company, not a chip company. Intel's overall contribution to the Linux kernel accounts for 10.5%. Intel's original intention of doing these open source works is definitely not to better benefit others, but to seek momentum like a Go master. This momentum is in the cloud computing era. Through cheaper open source software, cloud computing vendors can migrate from IBM's UINX+POWER to the more cost-effective Intel+Linux system. Now Intel occupies 95% of the server market. Intel's Xeon processors and cloud vendors such as Google Cloud, Amazon Cloud, and Alibaba Cloud are intertwined through these open source software. Now cloud vendors want to leave Intel behind (because of the need to reduce costs), but Intel has laid out its software ecosystem for so many years, and it will take a while to replace each cloud vendor. It still takes a lot of effort;

Another example is NVIDIA, which has open-sourced the CUDA compiler, CUDA toolkit, and development environment using the C/C++ programming language to the industry. Using this toolkit, you can develop, optimize, and deploy applications on GPUs. It also includes support for third-party tool chains (PyCUDA, Altime Hybridizer, OpenACC, OpenCL, Alea-GPU). At the same time, the CUDA ecosystem also includes a collection of many libraries, tools, and technologies. The CUDA library supports applications in a variety of fields such as linear algebra, image and video processing, deep learning, and graphics analysis.

CUDA open source also allows each user to bind NVDIA's GPU applications, so users can easily get what they want here. Instead of developing "wheels" from scratch, it saves time, effort and worry, and accelerates user deployment through ready-made libraries or ready-made application cases; with its powerful ecosystem and chip computing power, NVIDIA occupies more than 95% of the data center acceleration card market; it is through its entries in the field of AI that NVIDIA's stock price reached US$318.4 billion in 2021, surpassing Intel and becoming the number one in the IC industry;

The fourth level: "No core in the heart, no core in the hand"

If the first three levels still need to realize their own value by selling chips in the end, then the state of "no chips in the heart, no chips in the hands" is completely beyond this level, and the master of this level is Apple;

Apple also develops chips, but it does not develop chips for the sake of selling them. Instead, it develops chips to create a threshold that others cannot reach. Through chips, Apple endows its products with uniqueness and specialness that are different from other products. It also forms the faith of "Apple fans" and achieves the effect of "I am me, different fireworks".

The effect of Apple making chips and Apple developing a feature on iOS is the same. Customers are concerned about the final user experience of the product. The chip is just one part of the whole system, and it is a part that is difficult for others to imitate. Through vertical integration of upstream and downstream, unique product logic and product form, Apple provides users with something beyond their expectations, and users' psychological dependence on Apple arises spontaneously. Global fans "Apple fans" and "Apple worship" are born. At this moment, Apple's self-developed chips are the "holy grail" on the altar of "Apple worship".

Apple's chip research and development is like the sword demon "Dugu Qiubai". If it doesn't take action, it's fine. But once it does, it will shock the gods and ghosts. Last year, the self-developed processor M1 debuted. 5nm process, 16 billion transistors. The new unified architecture includes memory, CPU, GPU, and NPU. The GPU graphics processor in the M1 chip can provide 4K video editing capabilities this time; M1 surpasses INTEL by a large margin in terms of performance and battery life, which makes INTEL, the leader of the industry, feel embarrassed. Another huge advantage of the M1 chip is that it is compatible with IOS App. From now on, the architecture of iPhone, iPad, and MAC will be unified and unify the world!

In addition, there is Apple's mobile phone processor A series A14 Bionic, which is the most powerful mobile phone processor in the world today; its powerful performance ensures that your iPhone and iPad can run smoothly during the product cycle, and you don't have to worry about any problems such as lag. It is also the cornerstone of Apple's combination of software and hardware;

Another advantage of Apple that no one else can match is its strong self-developed capabilities. CPU, GPU, NPU, ISP, all are self-developed, which brings stronger performance and better power consumption; this is the barrier; it is the unique chip that gives Apple a transcendent state and also supports Apple's market value of 2 trillion US dollars;

If Apple is the unrivaled leader in the chip industry, it is hard to defeat the world; the only company that has the hope of reaching Apple's level in terms of chip development is Google. Google not only invited Hennessy and Patterson to help, but also made some amazing chips, such as TPU, which can be used in combination with Google's main TensorFlow. However, it has not yet been able to compete with NVIDIA, nor can it achieve Apple's vertical integration of chips and software, and its natural industry influence.

Final Thoughts

Boss Jia said that chip research and development is too difficult. Is there any shortcut for a rich man like Boss Jia? Money is not a problem!

Is this really possible? A turnkey chip solution from a design service company.

spec to chip; that is, Boss Jia tells the design service company what kind of chip he wants, and after paying the money, the design service company can deliver the customized chip to Boss Jia, it is that simple. The design service company is responsible for buying IP, assembling architecture, front-end design, back-end design, tape-out (find TSMC/SMIC), packaging and testing (find ASE/Changdian), and finally providing it to Boss Jia. The chips belong to Boss Jia. The design service company just collects hard-earned money;

Boss Jia was shocked. Isn’t this chip “surrogacy”?

Boss Jia tells the design service company what kind of child (chip) he wants. Boss Jia only needs to clearly describe the requirement of "boy/girl, what it looks like, what the eyes, nose, mouth, and ears look like" (chip SPEC), and the design service company will help you create it. However, surrogacy in the chip design industry is more reliable than the gossip that has been a hot topic in the entertainment industry a few days ago. As long as Boss Jia describes the functions, performance, power consumption, and cost indicators clearly, the design service company can give Boss Jia a qualified child (chip), without cheating anyone and guaranteeing your satisfaction.

Boss Jia asked, with such a good service, why don’t you use it more?

Because this kind of surrogacy (chip design service) is generally quite expensive. How expensive is it?

The man was silent, the woman wept. Boss Jia said he didn't care.

Boss Jia is a wealthy man, he can pay VeriSilicon, GUC, or even MTK to customize chips for him, but Boss Jia's power and status only go so far; if it were Microsoft CEO Satya Nadella, he could directly call "Su Ma" who works at AMD headquarters in Santa Clara and tell her that Microsoft XBOX needs a powerful SOC chip and must beat Sony's PS5. Su Ma answered the phone and said "it's a small matter", and directly took out AMD's signature ZEN2 and RDNA2 GPUs and customized a chip for Microsoft, which is the current XBOX SERIES X SOC chip;

This state is beyond the reach of the wealthy boss Jia; the happiness of the rich is so plain and boring;

*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.

Today is the 2607th content shared by "Semiconductor Industry Observer" for you, welcome to follow.

Latest articles about

■SiC giant, rebirth: how to predict the future?

■Apple chips may hit Qualcomm hard

■Chip cost per car: soaring to $1,000

■TSMC 2nm, important information

■Huang Renxun's latest views

■The risks of this type of chips that are promising have increased significantly!

■NPU, how to see it?

■Storage giants are abandoning DDR 4

■Intel, why?

■Nvidia will definitely be disrupted