The past and present of Labless
Abstract: Even for most professionals in the semiconductor industry, Labless should be an unfamiliar term. Let’s start with the familiar term Fabless. Over the past 30 years, the most significant industrial differentiation in the global semiconductor industry is the differentiation between WaferFab and Fabless.
In the 1960s and 1970s, the Silicon Valley in the United States gave birth to numerous great chip companies. In the early days when the supply chain was not complete, these chip companies could only use the IDM model, which means that they had to design, manufacture, package, and test everything by themselves. In the early stages of the semiconductor industry’s development, when profits were high, and cost was not sensitive, the IDM model undoubtedly played a positive role.
The industrial migration of IDM to industrial differentiation
In the mid-to-late 1980s, more chip companies entered the competition, and the several major drawbacks of the IDM model became evident.
1.Increased costs
a.Even for the production of a single chip product, the manufacturing and packaging must also be a complete production line with all the necessary equipment (photolithography machines, coating machines, etching machines, oxidation furnaces, cleaning machines, etc.). Inside the production line, the number of equipment in each production process needs to have the best matching scheme to achieve capacity optimization in each link. However, in fact, except for a few chip giants with huge sales, there are not many companies that can truly achieve this capacity optimization. The cyclical demand of single-product markets also causes fluctuations in production line loads, further exacerbating capacity waste. Therefore, production costs under the IDM model have always remained high.
2.Mismatched technology iteration speed
a.Moore’s Law predicts that the integration degree of integrated circuit chips will double, and their performance will improve every 18-24 months. Unfortunately, once the chip manufacturing production line is built and put into operation, there is not much room for further improvement. This means that except for a few types of chips with very long lifecycles, most production lines under the IDM model will be eliminated within two years due to their inability to keep up with their own chip design requirements. More advanced production lines must go through a series of technical stages such as technology route selection, equipment selection, production line debugging, and yield improvement, which easily causes a mismatch with their design department’s technical iteration. Intel, the leader of the old IDM model, is facing the risk of being completely surpassed by AMD’s long-term underdog due to the delay of its 7nm production line and is forced to turn to TSMC for foundry services, which is the inevitable result of this iteration mismatch.
3.Dramatic rise in talent costs
a.In the 1980s, more and more IDM chip companies entered the race, causing a severe shortage of semiconductor talent in Silicon Valley and a sharp rise in costs. The major companies were forced to move their packaging lines and wafer production lines out of Silicon Valley and to areas such as Taiwan and Singapore, which promoted the first industrial migration in the history of semiconductors and formed a physical separation of the design-manufacturing-packaging links under the IDM model.
b.Since the production and packaging lines can achieve physical separation from the design process, the contradiction between the production lines of more and more large factories and their own design requirements and technological iteration has become increasingly irreconcilable, providing a necessary growth environment for the later Fabless differentiation. In the late 1980s, professional third-party foundries led by TSMC and UMC emerged.
c.Independent third-party wafer foundries belong to the sharing concept and serve customers throughout the chip design industry, with inherent cost advantages. The “professional people do professional things” also forms a very good talent aggregation effect, and the speed of technological iteration far exceeds that of production lines under the IDM model. After more than a decade of evolution, in the decade from the late 20th century to the early 21st century, major IDM giants have all turned to Fabless mode, abandoning or partially abandoning their own production lines and embracing Fabless mode. The largest industrial differentiation pattern in semiconductor history has been formed.
China’s semiconductor industry-wide IDM model is a false proposition, the bottleneck does not lie in capital but talent
The special international environment, the implementation of the science and technology board and many other favorable factors have given birth to a large number of excellent semiconductor industry companies in China. However, the purpose of fundraising for a large number of chip design companies is to establish their own production lines and take the IDM model. In terms of the local needs of a single company, vertical integration with its own production line can guarantee capacity, customized production and enhance product competitiveness, which is undoubtedly a very reasonable use of funds. However, from a macro point of view, this kind of capital use is extremely risky, even “against the industry trend”. In addition to the cost and technology iteration mismatch risk we mentioned earlier, the bigger bottleneck in China is the talent.
The training of semiconductor talents is extremely long and expensive. Absolutely not by reading in the classroom can be completed, but also need expensive production line with and years of practical experience accumulation. It is said that in Belgium IMEC, it takes at least 5-10 years to train a truly qualified semiconductor talent, consuming an average of more than 5 million dollars of resources per person. From this perspective, the cultivation threshold of truly qualified semiconductor talents is comparable to that of a fighter pilot. If you add up the number of semiconductor-related graduates from China’s major universities and the number of talents independently trained by major companies, the number of semiconductor talents trained by China itself is only in the order of 10,000 per year. Yet the talent gap is widening.
In May 2018, figures showed that there were only 300,000 semiconductor employees in China at the time, but the actual demand reached 700,000, with a talent gap of 400,000. At that time, there were only 1,700 domestic chip design companies and less than 60 fabs.
Today, two and a half years later, China has more than 4,000 registered chip design companies and more than 110 fabs, with more fabs being planned and built. The number of companies has more than doubled, and if everyone is really doing something, it may take at least 1.5 million employees to support this industry scale. The talent gap is over a million.
A rough estimate, China’s current wafer manufacturing technicians only a small 30,000 to 40,000 people scale (including operators). Even the smallest IDM production line requires hundreds of people to participate in the production, the international IDM companies have a firm foothold in the scale of more than 10,000 people. Suppose there are 50 chip design companies in China to go IDM mode, digging up all the technical staff of existing fabs, may not be able to meet the demand for the number of personnel of these IDM companies. To achieve a firm foothold in the state, it requires more talent accumulation. In the talent pool is relatively limited, the background of the slow pace of talent training, can be expected to end up perhaps poaching each other, forced to lower the grade of talent, and ultimately the achievement of personal income soaring feast (the current income of semiconductor talent in China has exceeded overseas), but most companies can not do anything, the end will be a chicken feather.
A large number of chip design companies towards the vertically integrated IDM model, local seems logical and reasonable, does not mean that the macro viable. My pessimistic expectation is that after ten years, the number of IDM companies that can survive in China should be only a few, and mainly exist in a few areas that do not rely on advanced processes, but require very high process control, such as IGBT, power devices, RF and other fields. The rest of the companies must be forced to return to the Fabless model. In today’s semiconductor mania, we hope that we can enter the market with caution to avoid unnecessary business risks and waste of resources.
Industry segmentation is a historical inevitability, Labless will be born
In-HouseLab, the in-plant analysis laboratory, is a necessary auxiliary function department of major companies. It undertakes many functions such as product error correction, process monitoring, simulation, failure analysis, and R&D assistance. A good analysis lab plays the role of “doctor”, which can greatly accelerate the progress of product development, ensure the stability of production processes, and quickly find the root cause of product problems and give directions for technology iteration.
It sounds like Lab is a very important role for a chip product company. However, as mentioned above, just because the local business logic is reasonable, it does not mean that the macro is feasible. Let us do a little more analysis from several perspectives, such as cost, technology iteration, and talent.
1.Cost out of control
Early semiconductor chip process structure is relatively simple, and the size is only micron or submicron level, so a tungsten filament scanning electron microscope with a small amount of auxiliary equipment can complete most of the internal analysis business needs. However, as the device size enters the nano era, defects at the atomic level can cause device failure, and impurities at the ppb (parts per billion) level can cause serious drift in the electrical performance of the device, the analysis lab needs a wider variety of more sophisticated analytical instruments to have a chance to truly solve a problem. Can meet the current demand for a full set of advanced process analysis laboratory, only equipment investment of hundreds of millions of dollars into the scale, in addition to the wealthy fab, not many companies have the strength to undertake this infrastructure investment.
In addition, analytical instruments can not be as production-oriented equipment can do orderly production planning. The load of the analytical laboratory is more difficult to expect to control than the load of production-type equipment. When there is no condition on the production line, most analytical instruments are in an idle state, while once an accident occurs on the production line, it is imperative to resume production in a fraction of a second, regardless of how many analytical instruments are already owned by the in-plant demand side will expect a faster solution. This contradiction is almost irreconcilable.
2.The glass ceiling of the workplace and the degradation of the skills of the technical personnel in the plant have caused a hindrance to the iteration of analytical technology
In-plant analysis labs only focus on their own products, or even products on a fixed production line, every day. As mentioned earlier, once the production line is in mass production, it is not possible to do effective technical iteration on this line. Therefore, the skills of the analytical engineers serving this production line stop there. Such a team of analysts is unable to carry out effective analysis technology iteration, and their skills will not advance.
There is another secret that cannot be told. The status of analytical labs in major product companies is only a support function. The talents in this department, no matter how good they are, can hardly be promoted to the top decision-making level of the whole group. Therefore, it is difficult for the best analytical technical personnel to find the ultimate career home in a major product company. This is a “glass ceiling”.
A chip factory quality director once lamented to me, “Our factory’s analysis laboratory at the beginning of the investment is very considerable, and now there are hundreds of people, but somehow over the years only operators left, and even some projects we have equipment, but always do not do well, and even completely will not use”. This is partly due to the loss of top talent, but more importantly, the environment that binds the application scenario deprives talent of the necessary foundation for technology iteration, further causing the degradation of talent skills.
Teams that lose the ability to iterate on technology will be continuously eliminated by the never-ending technological progress. Without the support of a strong analytics team, more difficult problems will arise on the production line requiring the support of a stronger analytics team. This is a terrible vicious circle.
Various contradictions are slowly brewing and fermenting, slowly becoming irreconcilable, and even a paradox of macro data is emerging. Savor the words of the famous game theory expert, Yale University professor Martin Shubik. Shubik (M. Shubik) designed the classic case of the hundred-dollar bill auction (space limitations, not to repeat, the reader can search for themselves), perhaps now in the company’s factory analysis laboratory, and even a wider range of fields in full swing. And it seems to have reached a stage where bids are now over $100.
All the external environment is almost identical to the background of Fabless division. With further industry segmentation, Labless may be the inevitable end. Rational-minded people are waiting for the emergence of a “neutral third-party analyzer” that is strong enough to take the first step to end the $100 bill auction trap.
Who will be the winner when the opportunity arises?
A successful independent third-party analytics organization functions like a “comprehensive industrial hospital” with very high barriers. I have been in this field for more than 20 years and have been in business for 16 years, so I have some insights.
1.Hardware investment barriers
Investment of hundreds of millions of dollars is the first barrier in any case.
Insufficient investment in the laboratory, forced to do only fragmented analysis, or can only buy old models of equipment, resulting in long-term technical competitiveness is insufficient. There is another option, which is to engage in a joint venture with a partner who has the equipment, the cost, stability and timeliness are not controllable.
In addition, similar to the production line, the number of different analytical instruments is also concerned with the number of equipment ratios to achieve capacity optimization. If you buy only one piece of equipment of each kind, it will cause a huge waste of capacity in non-bottleneck link. However, the total investment scale must be increased several times to achieve the optimal ratio.
According to years of practical operational experience, the analytical platform with less than 100 million yuan of investment in equipment can only solve the problem of “availability” at most, while to really achieve capacity optimization, it is necessary to reach at least 500-1 billion yuan of equipment investment scale. Although the life cycle of analytical instruments than production equipment longer, but if you want to take into account the long-term sustainable technology iteration, the annual additional investment will also be very significant.
Various factors determine that small laboratories either sacrifice competitiveness or gross margin, both are not available. This is a painful climbing period, many companies stop here. Only third-party laboratories that have reached sufficient scale to enter the stage of capacity optimization will enter the stage of rising gross margins. At present, there are no more than five third-party chip analysis laboratories that can barely reach the state of capacity optimization in China.
2.Talent barrier
Independent third-party laboratories for the whole industry, the comprehensive ability of talent requirements are extremely high. After the formation of a certain scale, it is even less dependent on the personal heroism of a particular person. Very often, even for a specific case, often need to carry out “expert consultation”. Therefore, it is necessary to build a reasonable network of talents in order to be able to undertake more complex and difficult cases and serve more industrial fields.
Personal experience, the so-called experts from some large factories, in the commercial third-party platform may not be good. The reason is that his professional direction is too narrow, belonging to the “specialist”. Change a product, change an application scenario, he also needs to learn from scratch. To cultivate a “generalist” with a wide range of application scenarios, at least 5-10 years of practical experience is needed. Industry-wide “generalists” simply do not exist.
The bottleneck in the expansion of Concordia Hospital is the slow, if not irreducible, training of chief physicians. For independent third-party analysis platform, the bottleneck is also here. It is not only necessary to build a strong talent network, but also to form a benign talent cultivation mechanism and a standardized production management model to reduce the difficulty of talent cultivation.
3.Analysis technology IP barriers formed by strong independent R&D and technology iteration capabilities to enhance added value
The characteristic of technology products, especially consumer technology products, is that they will rapidly depreciate in value once they are separated from technological iteration. Only through continuous R&D iteration can we give new products higher added value and maintain high gross profit.
High-tech services are also a product that will also depreciate when divorced from technology iteration. 20 years ago, I was proud of my focused ion beam circuit modification technology, charging up to about $3000/hour. Today, 20 years later, the same technology node is available for 300 RMB/hour. But in these 20 years, the price of housing in Shanghai has increased 20 times. With housing prices as a reference, the loss of technology iteration of this “high-tech services”, depreciated 200 times.
The technology service industry relies heavily on human skills and cannot effectively reduce costs through mass production. In the context of increasingly expensive equipment and labor, the fees for analytics services should have continued to rise. However, in the past, the third-party analytical service market did not focus on the enhancement of added value, but fell into the strange circle of price war. Price decline, cost increase, there is bound to be a crossover point, no matter what, it is impossible to make a profit. This is the paradox of the business model, which will certainly end in a disastrous way with no winners.
Before 2015, China’s third-party semiconductor analysis services market was exceptionally bleak. Even several foreign giants also unspokenly generally use the old equipment eliminated from foreign headquarters to serve the mainland market. The overall service capability was exceptionally backward.
In 2016, Wintech Nano entered the Chinese semiconductor third-party analysis service market strongly, using the top international instruments and equipment in a clear-cut manner, and continuously increasing the added value of services through R&D iterations. In this way, it provoked an arms race with several overseas giants. Today, four years later, we see a new look in the entire Chinese third-party analytical services market, with several top-tier analytical services organizations on the mainland configured with almost no generation difference from overseas, and have comprehensively surpassed most major fabs’ in-plant laboratories. The entire mainland semiconductor industry is the ultimate beneficiary.
Third-party analytical laboratories are not only competing with their own friends, but also racing with their own customers’ in-factory laboratories. Not only to do it fast, but also to do it well. After continuous technical iteration, gradually open up the gap with the technical advantages of in-plant laboratories, will continue to gain customer recognition. Otherwise, we can only find scraps in the overflow of the customer’s in-plant laboratory to survive with difficulty.
4.Crisis of confidence and neutrality
Most of the third-party chip analysis laboratories are in contact with the customer’s more advanced or even being developed state of the future products. Historically, there have been members of the analysis organization plagiarized customer technical secrets, resulting in a serious crisis of confidence. This is the bottom line of basic business ethics. So customer trust in the information security system and long-term operation of the market reputation will be very important, never a one-day success.
And I do not see the instrumentation company as a side business of analytical services. Although there are indeed companies in the market to do so, but so far have not seen any of the precedent of success to do big. Selling equipment and selling services may seem complementary, but there is actually an essential paradox of sales motivation. In addition, most of the service providers of equipment companies are reluctant to purchase competitors’ equipment, which both loses neutrality and limits the core competency of the analytics organizations to learn from others.
Neutrality is neutrality
Neutral to both customers and suppliers. Wintech Nano’s philosophy is that a third-party analytical laboratory, which can only provide services, must not steal jobs from its own customers and suppliers. Therefore, neither the product nor the equipment can be touched. This is exactly the basic bottom line of TSMC’s business philosophy that enables it to be big.
When the general market environment tends to mature, the birth of an emerging industry is inevitable. This is a wave of long-term sustainable big market. There is no such thing as a winning era, only winners in a big era. The success of the technology industry is not a quick fix, not to mention the hot spots, the ultimate winners are bound to be those who are the most prepared, the most tenacious, the most cool-headed, the most able to adhere to the business bottom line of the long runners.
About the Author:
Xiaomin Li, B.S. from Peking University and M.S. from National University of Singapore (majoring in microelectronics), is currently the Chairman of Wintech Nano (Suzhou) Co., Ltd. and has applied for several invention patents and published more than 60 scientific papers and co-authored two books.
In 2001, he worked in Singapore Research Institute, and in 2004, he founded Wintech Nano (Singapore) Ltd. and in 2012, he founded Wintech Nano (Suzhou) Ltd. which is now a large-scale semiconductor chip analysis and testing service platform and auxiliary R&D center in China. The new concept of Labless was proposed to divest the “essential non-core” R&D link in the industry chain from the industry and become a new independent industry track. In May 2021, he was featured on the cover of the May issue of Singapore’s Time Magazine, becoming the youngest person to be featured on the cover of the publication since its inception, gaining appreciation and high recognition from the industry.