AMD CTO Mark Papermaster sat down with TSMC Co-COO Y.J. Mii to ask:
‘‘Have we run into the fundamental limits of physics or will innovation prevail again?”
(Can we keep shrinking transistors on chips?)
Thread 1/10 $TSM $AMD #semiconductors
2/10 TSMC’s Mii: Developing manufacturing processes in the 2nm and beyond age takes longer.
“In the past, indeed, it probably took 2-3 years to get things done, but right now because technology is more and more challenging, it takes more time, sometimes 5-years or even 7-years,” said Mii.
3/10 TSMC works side-by-side with customers like AMD to make sure design rules for new manufacturing processes meet client needs.
TSMC and other foundries provide design rules to customers, to ensure their chips are easier to manufacture.
4/10 TSMC starts work on next-generation processes years before they are in use
“We have 2 different technology development teams. They go one generation and the other group takes another generation, so we have more time and resources to work on the technology,” Mii said.
5/10 Working with customers is vital
“We have to work with our customers to make sure it really has value to our customers products,” said Mii.
The formal collaboration process is called ‘DTCO’ or Design Technology Co-Optimization.
6/10 TSMC and DTCO
“So, in our latest few generations of technology, we had a lot of collaboration with our customers and of course with our EDA vendors (Electronic Design Automation, Cadence, Synopsys) … to co-optimize the process and design,” said Mii.
Working together relieves the pressure of pushing the scaling and helps TSMC choose a less aggressive process, yet still get the same power, performance and density, he said.
7/10 New innovations are needed for every generation of technology
“…almost every generation of technology starting from 7-nanometer, you need to have new innovations,” Mii said. “We need innovation from materials, from processes, and in addition, we need innovation from design.”
A key innovation for 2nm is Gate-All-Around (GAA) transistors
8/10 TSMC 2nm moving from FinFET transistors to GAA
“This new generation of transistors really has a ‘gate all around’ the transistor channel so it really has better control of transistor performance and short channel effect, ... However, to get that technology done, it changed the whole process,” Mii said.
9/10 GAA, is “so complex that the DTCO is more important because the physics problems are a lot more complicated. So it’s extremely important to work together with design,” said Mii.
10/10 AMD’s Papermaster: “I feel you and I personally have embraced that approach. We brought our teams together very closely. We almost operate as one company and we respect, of course, the proprietary information from each side. But then on the areas where we intersect, where design meets technology, it’s an incredible collaboration.
And I think we’ve leveraged that, to great success, but never more than 2nm. 2nm brings on a whole other set of challenges.” $AMD $TSM #semiconductors
Through 2027, TSMC will see a “high teens” percent of revenue “from a very narrowly defined” AI application processor. - CEO
The yearly growth rate (CAGR) of AI at TSMC is about 50% “and we are confident that we can capture more opportunities in the future."
$TSM #semiconductors
3/15 more AI
Including other AI-related chips, networking and silicon for AI PCs, smartphones, etc., “if we lumpsum all the AI-related applications, actually, it’s quite a big amount for TSMC to grow.” - CEO
Reason for AI confidence: “our customers…asked TSMC to prepare the capacity, both in front-end and back-end…it’s in the early stage so far today – (but) we already see the increase, the momentum.”
$TSM #semiconductors #AI
3/9 What about Nvidia? It reports earnings on 2/21.
Based on 4th quarter guidance, Nvidia should end calendar year 2023 with revenue near US$58.82 billion, topping Intel and Samsung, but not TSMC.
Nvidia will win the 'standard' 2023 chip revenue title.
Note: The full year revenue estimate for Nvidia is based on its 1st quarter (Q1) to Q3 results, plus Q4 guidance. Simple math, nothing fancy.
Nvidia’s 4th quarter guidance is for its fiscal 2024, which ended Jan. 28, 2024. It can be confusing. Some companies follow a different fiscal year than calendar year, for accounting purposes. I’m comparing calendar year performance, which largely matches up. $NVDA $TSM $INTC #Samsung #semiconductors
Apple Silicon and TSMC - Thread 1/11
Apple Silicon lead Johny Srouji, and John Ternus, senior VP hardware engineering, talked about the marriage of hardware, Silicon, software, and TSMC during a CNBC interview.
$AAPL $TSM #semiconductors #semiconductor
2/11 Ternus:
“We’ve always had an incredible design team and we made these beautiful products, but they were constrained by what was available. And I think one of the most – if not the most – profound change at Apple, certainly in our products over the last 20-years, is how we now do so many of those technologies in house, and you know, top of the list of course is Silicon…”
3/11 Apple created the Silicon team to marry hardware, silicon, software all in one place, so instead of using off-the-shelf chips from other companies, they could create silicon precisely for Apple hardware and software.
Nvidia received a stern warning from US Commerce Secretary Raimondo on China export controls, media report:
“If you redesign a chip around a particular cut line that enables them to do AI, I’m going to control it the very next day,” she said, in a speech. 1/3 $NVDA #semiconductors
2/2 She urged Silicon Valley executives, US allies, others, to stop China from getting semiconductors and cutting-edge technologies vital to US national security, calling Beijing "the biggest threat we've ever had" and stressed "China is not our friend".
3/3 She also said her department needs more funding for AI export controls.
"Every day China wakes up trying to figure out how to do an end run around our export controls ... which means every minute of every day, we have to wake up tightening those controls and being more serious about enforcement with our allies," she said.
Q2 wafer shipments fell 9.6% to 2.916 million wafers from 3.227 million in the 1st quarter (Q1).
$TSM #semiconductors
3/x TSMC Q2 $TSM
Q2 revenue fell due to weak end demand and ongoing inventory adjustments among customers, TSMC said.
Q2 gross margin fell vs Q1 due to lower capacity utilization and higher electricity costs, partly offset by stricter cost controls & a favorable exchange rate.