China is hugely important for electronics in general and semiconductor in particular. You can't really appreciate it from the bubble of Silicon Valley, you have to go there.

For the second year, I attended SEMICON China in Shanghai (conveniently right next to the Cadence office...but inconveniently, the Cadence office is about to move). Last year, 80,000 attendees were expected and over 90,000 showed up. This year, over 100,000 were expected. It certainly felt like there were a lot of people there. To put 100,000 visitors in perspective, that is about the same number as attend MWC Barcelona, and a bit more than half the 180,000 who attend CES in Vegas.

There were 4,000 booths in the expo. The surprising thing, to me, about the expo is that it only fills about a third of the exhibition center and there are other shows running in parallel. In comparison, MWC Barcelona fills the eight hall convention center to overflowing, and CES basically fills the whole convention center and any hotel anywhere near.

The keynotes take place in the nearby hotel. This year they were:

• Zhang Suxin, chairman of Shanghai Huahong
• Timothy M. Archer, CEO, Lam Research
• Lip-Bu Tan, CEO, Cadence
• Akihisa Sekiguchi, CTO, Tokyo Electron Limited
• Choon Heung Lee, CEO, Jiangsu Changjiang Electronics Technology Co
• Sarah Cooper, GM, AWS Outcome Driven Engineering, Amazon
• Gary Dickerson, CEO, Applied Materials

Lung Chu

The keynote session kicked off with an introduction from Lung Chu. He pointed out that this was 31 years for SEMICON China. Whenever I hear anniversaries of long-lived conferences like this I wonder what they must have been like when they started, in 1988 in this case. In 1988 worldwide semiconductor revenues reached $477B which is a record, and growth in China was equally impressive.

Lung then introduced Ajit Manocha, the CEO of SEMI (world headquarters Milpitas).

Ajit Manocha

Ajit started off by pointing out how fast the world is changing. There are 30B devices connected today (that's IoT devices not just moble). Another 1.5M devices are attached every 10 minutes. There will be 80B devices in the next 5 years (by 2025). But only half the world is connected to the internet today. 

He predicted the global semiconductor market to be $490B in 2019 (up from 2018, which most analysts think won't happen due to declining DRAM prices). But even if 2019 is down a little, the direction is up and to the right. IBS numbers are $825B by 2027, and Ajit is sure we are on-track for a $1T semiconductor market by 2030.

Our biggest crisis is the talent shortage around the world. SEMI in 2017 pledged to put talent development as the highest priority around the word.

Ajit described SEMICON China as the "largest semiconductor show on the planet." (Maybe Elon Musk has plans for a bigger one on Mars?)

SEMI's mantra is "connect, collaborate, innovate." It is happening with AI, digital health, IoT, smart cities, smart transportation. It's a digital world we live in now.

There was then a quick pause for photography. That's all the keynote speakers above.

Zhang Souxin

Zhang Souxin spoke next. He is the chairman of Shanghai Huahong. The Huahong group was established in 1996 and built the first 200mm fab in China. They used it for a foundry business. They then built the first fully-automated 300mm fab in China. He said they have the only state-level R&D development center in China. They are now the world's 6th biggest foundry, with processes from 28nm to 0.5um. They are pushing to 14nm in their 300mm research fab.

Souxin's presentation was a fascinating Chinese perspective on the history of the semiconductor industry, starting with the basics. In some ways it was the perfect keynote for SEMICON China, showing China rising, but also pointing out the challenges ahead.

He started with the industrial revolution. In fact, going back even further:

1. In agricultural society, GDP was up 680% in 1820 years (arbitrarily starting in 1 AD I think), an average annual growth rate of 0.1%. Key technology, agriculture, of course.
2. In industrial society, GDP was up 720% in 130 years, a growth rate of 1.6%, driven by industrial production, initial steam-driven light industries like texties, then automation and electrification industries like cars and home appliances.
3. In the information society, GDP was up 630% in 48 years, a 3.9% growth rate, driven by computers and PCs.
4. In the internet society, GDP was up 220% in 12 years, a growth rate of 6.8%, driven primarily by mobile/internet and other technologies pulled along by it.
5. Today, we are in the intelligent society phase, after 2010, driven by mobile internet, IoT, and AI. Big data and cloud computing, AI, and more.

Souxin believes that with Moore's Law reaching its limit, specialty processes will become more important for emergent sectors such as IoT, and electrification of the automobile powertrain. Only a few foundies are pursuing the leading edge. Although in 2018 IDM revenue soared due to memory prices, in general, since 2000, fabless has grown faster than IDM.

Semiconductors are the top imported product in China, with $200B in 2013, rising to over $300B in 2018. The world's top 20 semiconductor companies have over half their market in China. It is the second largest equipment market in the world, and will become the largest this year. China's overall semiconductor market is growing at 8%, but domestic manufacturing is growing at 15% so the percentage of imports into the country will decline. It is expected to be over 20% by 2023.

My View

One of the fascinating things about the movement of semiconductor technology around the world was how unplanned it was. The story is not over as to how successful China will be in semiconductor, but they certainly are not going to fail due to lack of investment.

I think in the early days of semiconductors, it was all about semiconductor technology. If you had good technology, you could find the money to build a fab. If someone wanted to get into the semiconductor business, they would need to either develop their own semiconductor technology, or more likely kick-start things by licensing someone else's.

In the second phase, it was all about quality. The Japanese, in particular, completely embraced Edwards Deming's approach to quality and continuous improvement. Despite Edwards being an American, the US did not. Somewhere I still have my signed copy of his book Out of the Crisis from when I attended a 2-day seminar when he was already in his mid-80s.

In the third phase, it was all about money. A fab was so expensive that it was hard to raise enough money to build one, and the economical size was beyond the total business of most semiconductor companies. Every IDM in the world except Intel and Samsung dropped out. All the memory manufacturers build their own memory fabs too—the smallest increment of memory capacity is an entire fab. in effect, as the fabless model took hold, process technology was easy enough to obtain since you just went to a foundry and got it by ordering wafers. We are now at the stage where only a few countries can afford to build a leading-edge fab.

I think there are two likely scenarios, over a 5-year or 10-year timeframe. One scenario is that China gets the experience and has manufacturing technology in the three key areas (logic, DRAM, 3DNAND) that domestic manufacturers can use the products and be competitive, initially perhaps only in China. But companies like Xiaomi, Oppo, and Vivo have shown that you can be in the top 5 worldwide serving mainly only China. Lenovo has shown that a Chinese company can be globally successful, albeit starting with a big boost by taking over IBM's laptop business.

If Chinese semiconductor manufacturers get to be good enough that domestic manufacturers can use their output, there will be tremendous pressure to use it. The Chinese market is then so big that they will automatically develop the manufacturing experience, and have self-sustaining cash to invest in future process technology.

The other scenario is that what happens to this huge push into semiconductor is what happened last time. It doesn't work out. For example, if China can't get DRAM to have reasonable quality then the Oppos and Vivos will continue to have to source DRAM from non-Chinese companies. But without the volume, there is no way to get onto the learning curve to world-class yield.

Other Keynotes

I'll write about a couple of the other keynotes in the next couple of weeks.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.