What is a "specialty technology"?

Kevin Zhang, the VP of business development, told us at the recent TSMC Technology Symposium:

If Yuh-Jier Mii doesn't talk about it, then it's a specialty technology

Yuh-Jier Mii is the head of technology development for the mainline digital processes: 7nm, 5nm, 3nm, and beyond. He had spoken earlier that morning. To find out what he did talk about, see my post from last week TSMC Technology Roadmap.

TSMC Has Never Closed a Fab

Kevin explained that specialty technology runs in Fabs 2, 3, 5, 6, 8, 10, 11, 12, 14, 15, 16. I don't know why, but TSMC doesn't have a fab 4, 7, or 9. Here is TSMC's list of all TSMC fabs.

That's a lot of fabs, but here's an amazing thing. Cheng-Ming Lin pointed out last year that TSMC has never closed a fab. Fab 1 was leased in the very early days, and they did give it back. But that's it. That was at last year's Technology Symposium which I reported on in my post TSMC's Fab Plans, and More. I assume the missing fab numbers got merged into adjacent fabs.

The strategy is to run mainline digital when a fab is new, and then find more lines of business to run in it as the volume of digital wafers moves to the next fab over the following years.

The fabless/foundry business model that TSMC pioneered has several advantages over what we just called the "semiconductor business model" and now say "integrated device manufacturer" or IDM. In the old model, where a semiconductor company had fabs, the whole mentality was to fill the fab exactly with lines of business. The company needed just the right amount of business to fill its fabs: not enough, and the excess depreciation was a financial burden, too much and they were on allocation and turning business away. When a fab got too far off the leading edge, there might not be enough business to run in it, and the fab would be closed. Or, in the days when many companies had both memory and logic businesses, they would run memory and then convert the fab to logic, and then eventually close it.

TSMC finds new lines of specialty technologies, finds new customers, and keeps the fabs going. Things like CMOS image sensors (CIS) or MEMS (micro-electrical-mechanical-systems such as pressure sensors). Adding new options to non-leading edge processes such as novel non-volatile memories.

It's almost as if Henry Ford's original factory for the Model T was still running, but now making turbine blades for the aerospace industry.

5G RF

As was clear from Kevin's presentation, the hottest new area is adding modules to their processes for 5G RF design. In fact, in CEO CC Wei's opening presentation, he invited Qualcomm's CTO James Thomson on stage to explain it.

In his words:

millimeter wave is a completely different system, but hard to use

He also explained the massive multi-user MIMO in the traditional bands [known as sub-6GHz], requiring 256 antennas with up to 64 simultaneous digital streams, which can essentially "deliver hot spots in 3D space." On the phone, there is an antenna at each corner of the phone and can deliver up to four data streams, and then take another 15 phones and deliver their four data streams. Plus, economically, they have to reuse 4G towers.

But, as he continued: "mmWave is hard to use due to path loss being a hundred times more than traditional bands" [this means almost anything will block it, including walls, windows, hands, and more than about 300m of air, since oxygen is a big absorber]. Innovations are required to overcome the limitations. One interesting datapoint James said:

In fact Steve Mollenkopf, CEO of Qualcomm, worked on mmWave in grad school and so was an initial skeptic.

Anyway, this isn't the point to go into depth on the whole 5G business, that's for another day, but just take away that 5G requires a lot of innovation in RF.

Growth in Specialty

Kevin said that specialty has had a growth since 2013 (5.7M wafers) to 2018 (11.1M wafers). That's a CAGR of 14.3%. Their investment has had a CAGR of 10% over the period. They've moved MEMS from 8" to 12" manufacturing. They've moved CIS from 0.18um to 0.08um. They've switched from conventional [floating gate] to emerging memory.

He said he would go into detail on a few key highlights:

• ULP (ultra-low power) for IoT
• Advanced RF for 5G
• Embedded flash and new memories
• Power management
• MEMS

Their approach is to build integrated specialty platforms, layering options onto a basic logic technology at the bottom and a common design platform to share the investment. So layer on modules like NVM, BCD, HV, sensors.

Some Notes

Here are a few somewhat random notes since it was like drinking from a firehose. There are so many specialty processes, so many options, and so many buzzwords, it's hard to keep up.

Their low-power process portfolio includes 0.18um ELL, 90nm ULP, 55ULP, 40ULP, 22ULP, 22ULL, and 12FFC+_ULL. Voltages from 0.75V in 55ULP down to 0.5V in 12FFC+_ULL.

5G RF solutions:

• For sub-6GHz,16FFC RF with RF SOI
• For mmWave, 16FFC RF with 28/22 RF

I'm not an RF guy, but in case you are, there are key improvements with substantial reduction in insertion loss, inductor Q enhancement with elevated ultra-thick-metal. RF calculator and aging model for mmWave power-amplifier design. There is also an advanced RF with FinFET 16FFC with PDK and SPICE scheduled for 1Q 2020.

In the embedded non-volatile memory, there has been a problem that flash type memory doesn't scale beyond 28nm, and even there has a cost problem requiring 18 additional masks. TSMC is moving forward with MRAM and RRAM:

• 22ULL MRAM risk production this year
• 40RRAM risk production in 1H'19
• 28/22 RRAM risk production 2H'19
• eFlash alternative for IoT
• 22MRAM risk production in 2H'18

Or, as Kevin summed it up:

it’s been emerging memory for many years and now I’m proud to say it’s emerged

A few more highlights:

• Advanced-Power IC solutions: Low-power 40nm ULP logic, BCD_ low Ron NLDMOS up to 24V. Incorporate RRAM fo NVM.
• Display evolution: N80HV/N55HV, moving to N40/N28V. Scaling display technology from 0.18um down to 28nm.
• World's first 28HPC+ with 40HV WoW offers best power and form factor with quick TTM. Also provide monolith N28HV technology in Q1 2020.
• Camera is important, especially in the Asia market. Pixel size driving to 0.8um this year, and 0.7um next year.
• Infrared sensing important for security and ADAS. Achieved 940nm QE >45% (compared to silicon).
• MEMS pressure sensor with 5-10cm sensitivity in altitude change.

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