outlook丨de-sinicization of advanced packaging is counterproductive
2024-08-31
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
◇advanced packaging technology plays a key role in the ai chips required for the burgeoning ai technology revolution
◇the reality of the current global semiconductor industry chain is that more than 60% of chips need to be shipped to china and more than 90% of chips need to be shipped to asia for packaging and testing before they can be sold in the global market, including western countries.
◇in addition to malaysia, japan, singapore, vietnam, the philippines, india, etc. are also important destinations for the united states to deploy overseas semiconductor manufacturing bases
◇at present, china and the united states are at the same starting line in the field of advanced packaging. the advantages that the united states has accumulated in the field of semiconductor miniaturization over the past few decades can no longer be used to restrict china in the field of advanced packaging.
text | tan xiaojian
staff members are testing a laser radar chip in the packaging and testing public service platform workshop of suzhou institute of optoelectronics technology (photo taken on june 2, 2024) photo by li ga/this magazine
not long ago, the u.s. department of commerce issued an announcement, announcing the launch of an innovation investment fund with an initial total of us$1.6 billion to realize and accelerate the implementation and upgrading of the u.s. semiconductor advanced packaging industry. as the highest priority item in the "chips and science act" dedicated to semiconductor technology research and development funding plan, why is advanced packaging technology so important? what is the purpose of the united states' layout in related fields?
the rise of advanced packaging
when discussing the development of chip technology, "moore's law" is a term that cannot be avoided. its basic meaning is that the number of transistors that can be accommodated on a unit integrated circuit doubles every 18 to 24 months.
"moore's law" is an empirical law proposed in 1965 by gordon moore, director of the research and development laboratory of fairchild semiconductor corporation (the predecessor of intel). it reflects the digital industry's demand for the ever-evolving chip technology and is also a business strategy for us chip companies to maintain their leading position.
on the one hand, the continuous improvement of chip performance has promoted the sustained growth of the u.s. digital industry. after tasting the market benefits brought by digital technology, technology giants will in turn invest heavily to support u.s. chip companies in developing the next generation of chips, allowing u.s. chip companies to invest heavily in technological improvement and innovation.
on the other hand, the continuous updating of technology has also led to a continuous decline in chip prices with stagnant technical indicators. the united states has suppressed competitors of u.s. chip companies by releasing such price reduction expectations to the market. the more the latter lag behind, the greater the losses will be, and they may even go bankrupt due to insufficient income to cover expenses.
the combined effect of these two factors has resulted in the "stronger gets stronger" in the semiconductor field. "moore's law" is not only an industrial law, but also helps the united states maintain its dominant position in the field of information technology.
the reason why the united states is so confident that it can become the main winner under "moore's law" in the long run is that it has seen the commercial technology development path of "semiconductor miniaturization" and believes that this path can be carried out for decades - the design and manufacturing of chips only needs to continuously reduce the size of the circuit and use shorter wavelength light for lithography. in this way, chip performance can be improved in a relatively stable manner and at a lower cost, and maximum profit can be obtained with each process reduction, thus achieving long-term development.
however, around 2010, as "semiconductor miniaturization" continued to approach its physical limits, the economic demand and business strategies supporting "moore's law" began to waver.
on the one hand, the cost of developing miniaturized components is rising. finfet technology was introduced when developing the 16nm process, extreme ultraviolet (euv) lithography technology was introduced when developing the 7nm process, and now all-around gate transistor (gaa) technology is introduced when developing processes below 5nm... behind every technological innovation, there are a lot of additional technical breakthroughs. not only is the r&d investment getting higher and higher, the frequency of additional technical breakthroughs is also getting more and more frequent, and the profit margin of new process chips is constantly declining.
on the other hand, the chip performance improvement brought by semiconductor miniaturization has become increasingly "useless". according to tsmc's technology roadmap, the 3nm chip has a 1.7 times higher transistor logic density than the 5nm chip, but the performance has only increased by 11%, which is a significant reduction compared to the nearly 50% performance improvement of each previous miniaturization.
this has led chip manufacturers to carefully consider r&d investment in lower-process chips and seek technology development paths beyond "moore's law." it is in this context that advanced packaging technology has begun to receive more and more attention.
if semiconductor miniaturization is focusing on "integration within the chip", then advanced packaging has set its sights on "integration between chips" - by adopting a special packaging method to improve the integration of chips and external components, multiple chips are integrated into an organic whole to play the computing function originally performed by a single processor chip, thereby achieving the goal of continuing to improve the performance of computing units beyond "moore's law". this is called the "super moore's route" by the semiconductor industry. at the same time, research and development on semiconductor miniaturization is still continuing, which is called the "deep moore's route", but it is no longer the only path to improve semiconductor performance as in the past, and its breakthrough speed is gradually slowing down. this new semiconductor development period, which is significantly different from the "moore's law" era that relied solely on semiconductor miniaturization in the past, is called the "post-moore era."
it is worth mentioning that advanced packaging technology plays a key role in the ai chips required for the raging ai technology revolution. for example, the nvidia h100 chip, which is currently shining in the field of artificial intelligence, uses the cowos advanced packaging technology developed by tsmc. the principle is to insert an intermediate silicon carrier between the computing chip and the memory chip, organically connect and package them together, so that the speed at which the chip accesses external data is significantly improved. this high-speed connection between chips achieved through advanced packaging is named high-bandwidth memory (hbm) technology. the hbm bus also has an external interface that can be used to connect more external chips in series. its theoretical transmission speed can reach up to 450gb per second, which is dozens of times the previous communication speed between the chip and external data, and there is still a lot of room for improvement.
it can be said that advanced packaging technology is guarding the intersection of the "artificial intelligence era" and the "post-moore era". without advanced packaging technology, artificial intelligence chips would not have made such great progress, and the development of semiconductor technology would also slow down.
america's sense of crisis
traditional packaging and testing, as the last process before a chip leaves the factory, has always been at the end of the industry chain. it not only has meager profits, but has also been regarded as a low-tech and labor-intensive industry. in the 1970s and 1980s, when the united states moved part of its semiconductor industry to asia under the "fabless" and "offshore outsourcing" business strategies, the chip packaging and testing process was almost the first to be outsourced, first to south korea and taiwan, china, and then to mainland china, as well as southeast asian countries such as malaysia and vietnam.
in 1978, with the strong support of the state, chinese companies introduced 5-micron chip packaging and testing production lines from toshiba in japan, which opened the development of my country's chip packaging and testing industry. today, some chip packaging and testing manufacturers in the mainland have developed into international packaging and testing giants with annual revenues of hundreds of billions of yuan, accounting for nearly 20% of the global chip packaging and testing market share, second only to taiwanese companies such as ase. at present, companies on both sides of the taiwan strait together account for more than 60% of the global chip packaging and testing market share. however, american, european, japanese and korean companies have a relatively weak presence in the field of chip packaging and testing. only amkor technology in the united states accounts for about 14% of the global market share, and the rest are all below 1%. amkor technology's packaging and testing plants are also built in asian countries, and only the technical research and development department is retained in the united states.
the reality of the current global semiconductor industry chain is that the united states has an advantage in the upstream of the industry chain, such as chip design and manufacturing tools, while china has firmly occupied the packaging and testing link at the end of the industry chain. more than 60% of chips need to be shipped to china, and more than 90% of chips need to be shipped to asia for packaging and testing before they can be sold in the global market, including western countries. as advanced packaging technology plays an increasingly important role in the "post-moore era" and "artificial intelligence era", such an industrial structure has aroused the vigilance of the united states.
in november 2023, the first r&d investment project of the us chips and science act was invested in advanced packaging technology. the act specifically allocated $3 billion to fund chip packaging companies in the united states, and the program was named the "national advanced packaging manufacturing program." on july 9, 2024, the us department of commerce issued an announcement announcing the launch of the first batch of r&d funding awards totaling $1.6 billion under the program, covering five sub-fields of advanced packaging technology, with each r&d innovation award up to $150 million.
the us's pursuit of "strategic autonomy" is difficult to achieve
driven by the political circles, american companies have stepped up their layout in the advanced packaging industry.
in december 2023, amkor technology announced that it would invest $2 billion to build an advanced packaging and testing plant in peoria, arizona. apple immediately expressed its support, saying that apple would become the first and largest customer of the packaging and testing plant. intel has also been continuously working on technology research and development, independently developing foveros advanced packaging technology, and announced a new chip architecture to adapt to this technology.
american companies are also entering overseas markets. in august 2023, intel announced that it would invest heavily to build its first overseas advanced packaging and testing plant based on its technology in penang, malaysia. in december 2023, nvidia announced that it would build a cloud computing center in johor, malaysia. with the influx of american companies such as amd and micron, the development of malaysia's chip packaging and testing industry has accelerated. in addition to malaysia, japan, singapore, vietnam, the philippines, india, etc. are also important destinations for the united states to deploy overseas semiconductor manufacturing bases.
the united states’ goal of seeking “strategic autonomy” through this may be difficult to achieve.
first, at present, affected by factors such as the high cost of building factories in the united states, the progress of tsmc's arizona factory and other key projects funded by the us "chips and science act" has been delayed to varying degrees or even suspended; foreign companies investing and doing business in the united states also face "incompatibility" in terms of business culture and political manipulation.
second, like the united states, countries and regions such as japan, southeast asia, and south asia lack a sufficient number of highly skilled workers, and many upstream industrial products and components in the chip manufacturing field have to be imported from mainland china, taiwan and other places.
what this reflects are the various unique advantages accumulated over the years through long-term investment in industries, science and technology, education and other fields. these include rich and complete local industrial resources, relatively low overall manufacturing costs, a large technical population, etc. these have made china an indispensable and irreplaceable presence in any advanced manufacturing field.
at present, china and the united states are at the same starting line in the field of advanced packaging. the advantages that the united states has accumulated in the field of semiconductor miniaturization over the past few decades can no longer be used to restrict china in the field of advanced packaging. the united states' promotion of "de-sinicization" in the field of advanced packaging will not only not restrict my country's scientific and technological development, but will instead inspire my country to make continuous breakthroughs and form stronger competitiveness in the global semiconductor industry chain and the global market.
(the author is an associate researcher at the institute of science, technology and cybersecurity of the china institutes of contemporary international relations and the head of the artificial intelligence project)
(outlook, 2024, no. 36)
