This week was the GLOBALFOUNDRIES Technology Conference, GTC 2019, in Santa Clara. It was the 10th anniversary of GF, which was created on March 4, 2009. it is also almost exactly one year since what GF calls "the pivot", where they decided to abandon the "single-digit nanometer wars".

History: Founding to the Pivot

Here is a short summary of where GF is today and how they got here. GF was originally created in 2009 by spinning out the manufacturing from AMD, which came with Fab 1 in Dresden, Germany. They acquired foundry Chartered Semiconductor later that year, along with several fabs located in Singapore. In 2015, they acquired IBM's semiconductor business, along with fabs in Burlington, Vermont and East Fishkill, New York. They built a green-field fab (fab 8) in Malta, New York. After the pivot last year, they rationalized their manufacturing facilities. They sold their smaller fabs, mostly used for MEMS, in Singapore to Vanguard International Semiconductor (VIS). They agreed to sell their East Fishkill fab (fab 10) to ON semiconductor, with a 3-year transition plan to gradually move the business they make there into Fab 8, along with the trusted foundry certification (so they can manufacture for the US government). Fab 9 (Burlington) is also a certified trusted foundry. Last year, they spun out their ASIC design business into a subsidiary Avera, and then this year sold Avera to Marvell.

So that leaves them with:

• Fab 1, 300mm in Dresden
• Fab 7, 300mm in Singapore
• Fab 8, 300mm in Malta
• Fab 9, 200mm in Burlington
• "GigaFab" which I think are the remaining 200mm fabs that they kept in Singapore

After the pivot, GF abandoned plans for processes below 12nm, in particular, they stopped the development of a 7nm process and returned their EUV steppers. They then took the $20M per month they were investing in 7nm and used some of it to invest in specialty processes.

Is it working? In the press and analyst lunch, Tom said that:

for the first time in ten years we have positive cash-flow. And not just a dollar. We are $6B in revenue and had a positive cash-flow of $550M. That doesn't include the transactions.

He was asked if this was public information. "It is now," he said. When he was asked if they would do something on a more formal basis to make the numbers public (as a private company they don't have to) he said "first and foremost it is important that our customers know how we are doing.”

The big motivation for the pivot was that they were not at a scale to be competitive at the leading edge. "Our competition has a CapEx budget bigger than our revenue." They also realized that by giving up on processes below 12nm, they were giving up 25% of the market, but keeping 75% where, they felt, with a more focused investment they would be very competitive. Or, as Tom put it:

We felt the world didn't need 3 companies going to 7nm and we weren't at a scale to get a return on that. It just didn’t make sense, and we needed to convince  ourselves we could be a winner as the world’s biggest and best specialty foundry. I have more conviction now. Moore’s Law scaling is dyiing, and 12nm is the last platform to integrate features on.

That last sentence is an acknowledgment that increasingly integration using chiplet technologies is the way of the future. Based on many of the presentations at the recent HOT CHIPS, which I described in my post HOT CHIPS: Chipletifying Designs, it certainly looks that way for many of the highest performance designs. 

GF Today

In his opening keynote in the main session, titled The Future of Innovation, Tom put the business in perspective:

• Global GDP is $85T
• Electronics is $2T
• Semiconductor is $475B
• Foundry is $63B
• There are just 5 foundries at scale supporting global GDP

Tom focused in on what he calls AIM (auto, industrial, and multi-market), computing and wired infrastructure, and mobile and wireless infrastructure. Later, in his presentation, Bami Bastani sized this 3 markets (at 12nm and above) as $24B for AIM, $15B for computing and wired, and $8B for mobile and wireless, adding up to the $47B representing 75% of the $63B foundry market.

In fact just that morning, GF had put out a press release announcing that they had hired Mike Hogan to lead the AIM business unit. He joins Bami Bastani who heads up the computing and wired business uint, and Mike Mendicino who heads up the mobile and wireless business unit.

Tom showed that GF can grow 1.4X their current size on the current footprint of the fabs I listed above.They can grow wafer starts from 2.3M to 3M and revenue from $6B to $8B without needing to build any fabs, just move in more equipment. The diagram below shows the headroom:

Instead of innovating by simply scaling in a race to keep up with Moore's Law, GF is doing that Tom calls "platform innovation", which is combining their basic process platforms with special application features and specialized IP. He multiplied the numbers out to get to 10s of thousands of specialized application solutions.

He went on to give some examples:

• 12nm FinFET with LV SRAM bit cell and 2.5D interposer with an AI reference package to deliver 12LP+ datacenter AI
• 45nm with power-amplifier FET, high-performance switch, and NLA with 1.0V digital to give 45RFSOI 5G mmWave REM
• 22FDX with RF mmWave AutoPro to give 22FDX-RF ADAS Radar

Later, in his presentation, Gregg Bartlett filled out the details:

Mike Cadigan gave some more statistics on the scale of the operation to support all these combinations and the designs that result:

• Use 2000 mask sets per year
• Have over 100 ecosystem partners
• Deliver over 300 PDK releases per year
• And, as I already said, 2M wafers per year

GTC in Rest of World

If you are not in the US, there are three upcoming GTC events:

• October 11 in Munich (in the Park Hilton)
• October 24 in Shanghai (in the Kerry Hotel)
• November 5 in Taiwan (in the Sheraton Hsinchu)