The electronics industry continues to struggle with supply chain problems that started with the ongoing US-China trade war and were exacerbated by the lingering COVID pandemic. Barbara Jorgensen, editor of EPS News, has compiled a series of articles (appearing across sister publications EPS News, Electronic Products, and EE Times) that illuminate what is happening in the global supply chain and offer advice on how to adjust to all the changes. Links to those stories are below.
In April of 2021 Google announced it would be making custom systems-on-chip (SoCs) for its own server farms. The reason? To boost performance and efficiency, according to Amin Vahday, Google’s VP of Systems Infrastructure. Relying on individual chips on a PC board was getting just too slow.
The SoC decision makes perfect sense and is in line with Google’s DIY approach for staying on the cutting edge of digital technology. A successful earlier case in point is the Tensor Processing Units (TPUs), developed in-house and introduced in 2016. These robust custom-made application-specific integrated circuits (ASICs) were used to speed up machine learning (ML) workloads and paved the way for a host of new services.
Overlapping the SoC news, Google announced it had hired veteran Intel chip designer Uri Frank to be Google’s new VP of engineering for server chip design.
“We buy where it makes sense, build it ourselves where we have to, and aim to build ecosystems that benefit the entire industry,” said Vahday.
But the supply chain had other plans. Chips have been in short supply since January and are expected to remain scarce through 2022. Fabless chip companies are clamoring for the attention of independent foundries. The semiconductor industry plans to add capacity – that will come online in two to three years.
THE FAANG business model
Google isn’t the only Tech Titan making a move to design and build its own hardware.
In March 2019, Facebook announced it was developing new ASICs in-house optimized for AI inference and video transcoding.
In June 2020, Apple announced it was parting ways with Intel and would design its own SoC, the M1, for Macintosh computers.
In late 2020, Microsoft revealed it was working on an in-house processor using ARM designs for servers that run cloud services in its own datacenters. The company is reportedly considering using another chip to power its line of Surface personal computers.
In early 2021, Amazon Web Services (AWS) announced its custom-built Graviton2 processors using 64-bit Arm Neoverse cores.
The moves by Microsoft and Apple were prompted by the risk of an overreliance on Intel for next-generation chip technology. Intel has lost its dominant position in cutting-edge technology in recent years, starting with delays of its 10-nm node chips and then 7 nm, and now 5 nm and potentially 3-nm delays.
Other born-digital Titans that have built their empires on cloud infrastructure rely on Google, Microsoft, and AWS to provide server technology that can quickly scale new services. For instance, Netflix relies on AWS for compute and storage to keep its movies rolling.
Supply chain choke point
This group of global cutting-edge, cloud-centric companies, sometimes called the FAANGs – for Facebook, Amazon, Apple, Netflix, and Google – operate in a rarified environment given their size, market ubiquity, and financial power. They design highly sophisticated customized chips, ASICs, and SOCs that set the bar for other lesser companies to try to match.
However, when it comes to the actual manufacturing of the silicon wafers, they still have to farm out production to an independent fab, just like the non-FAANGs. That means contracting with a foundry, which is likely to be a pure-play foundry on the leading edge of chip technology and doesn’t design its own chips.
According to IC Insights, pure-play foundries are forecast to account for over 80 percent of total foundry sales in 2020. Through 2024, pure-play foundry sales are forecast to reach a compound annual growth rate (CAGR) of 9.8 percent. In 2020, the global foundry revenue was $85 billion. Taiwan accounted for 63 percent of the global foundry market, with TSMC accounting for 54 percent, according to TrendForce. Korea’s Samsung is second at 17 percent followed by Taiwan’s UMC and US-based GlobalFoundries at 7 percent each. China’s SMIC produces 5 percent of the world’s foundry chips. In 2021, TrendForce projects the total foundry business to increase by about $10 billion to $94.6 billion.
The top foundries are leading the way in chip technology, with leading-edge geometries currently at 7-nm and 5-nm range with 3 nm just around the corner. This is where the FAANGs play. While they don’t broadcast which foundry they use, pundits point to either TSMC or Samsung. Google won’t say which foundry it is planning to use for its game-changing SoCs, but speculation in the industry is that it will probably be TSMC.
Who gets the capacity?
The challenge today for chip manufacturing is that demand for wafer starts is high and capacity is tight. An unexpectedly strong fourth quarter in 2020, driven by high demand for automotive and consumer electronics, resulted in chip shortages throughout 2021 and fab capacity being booked out for the foreseeable future. The cleanroom fire at the Renesas fab in January didn’t help the cause.
(Source: VLSI Research)
With chip demand accelerating, the only short-term option for fabs to meet the rising demand is to increase capacity utilization, with some individual fabs running as high as between 90-100 percent. The industry has been steadily increasing overall fab utilization over the past two years and is estimated to increase utilization even more during most of 2021 to meet demand. Higher fab utilization will increase chip output and allow the industry to better meet the increased demand in the market.
It’s typical that when chip demand is high, as it is today, fabs will typically run at 80 percent or higher capacity utilization, according to the Semiconductor Industry Association (SIA). For the last two years, utilization rates have been increasing and were projected to climb above 95 percent by Q3 2021, according to VLSI Research.
The end-to-end process for producing a finished chip – from ramp-up to volume production to assembly, test and packaging – can take anywhere from 26 to 50 weeks, according to SIA. The more complex the chip design, then more time is required for the manufacturing process.
Implications for the FAANGS
In the short-term, the semiconductor industry is doing all it possibly can to increase utilization and meet increased demand in the auto sector and more broadly for all customers including the FAANGs. Foundries are in a position to pick winners and losers by choosing which chips to produce.
For Google to get its SoCs produced so its server farms can outperform the competition, comes down to just two fabs: TSMC and Samsung, said Dale Ford, chief analyst, Electronic Components Industry Association. “The types of SoCs that Google will use as co-processors will be based on bleeding edge process geometry technologies,” he said. “They have to choose one or the other.”
As for Google’s TPU, it is on a schedule to be at 10 nm, which means either TSMC or Samsung is making it, said Brian Santo, editor-in-chief of EETimes. “TSMC is not only the best right now, is very likely to keep being the best for the next few years. So, when you go to them, you have a solid expectation that doing so is safe for your future products too,” he said.
But the fabs have choices too — and constraints — especially in a time of high demand and near-maximum capacity. Google and the other FAANGs have an advantage that chips for data-center servers is a high growth sector. “This will make the server market segment attractive to the fabs as they look at longer term growth opportunities,” said Ford.
Still, foundries are in a position where they have to honor commitments that they have already made to customers. Trying to jump to the head of the line is difficult if not impossible.
But if the FAANGs have done proper planning and forecasting and already have their orders in process they should be okay, Ford said. However, he warns that there won’t be any opportunities to scale up their orders in the near term.
And over the longer term, companies that can commit to higher volume and higher value production — like smartphones — will receive priority in getting fab capacity. “So, the FAANGs will still sit behind the ultra-high volume smart-phone customers,” said Ford.
What’s over the horizon?
It will take a few years to resolve the capacity constraints. TSMC and Samsung both have plans to build new fabs in the U.S., but the plans are in the early stages. In the meantime, geopolitics could play a role as China is growing increasingly vocal about its desire to bring Taiwan back into the fold.
President Joe Biden’s administration stated in early 2021 that the U.S. was committed to supporting an independent Taiwan. However, China is harassing the nation with flyovers of Chinese warplanes. In March, Adm. Philip S. Davidson, the U.S. military commander for the Indo-Pacific region, stated that China could try to reclaim Taiwan by force within the next six years.
In April, China flew a large number of military jets into Taiwan’s air defense zone. In response, U.S. Admiral John Aquilino, head of the Pentagon’s Indo-Pacific command, warned that a Chinese invasion of Taiwan “is much closer to us than most think”.
If he’s right, concerns over Google’s SoCs and the FAANGs will likely take a backburner.
About the author
Bruce Rayner is principal of Locust Content Services, specializing in crafting effective thought leadership content that communicates clients’ insight and expertise.
The EPS News supply chain package:
Navigating a Path to the Supply Chain’s ‘New Normal’ (EPS News), by Bruce Rayner.
It’s fair to say that the supply chain in 2021 didn’t evolve as expected. A chip shortage at the beginning of the year has only gotten worse; global logistics are hopelessly backlogged; prices are on the rise and a new variant of Covid-19 is emerging. Despite this, the U.S. manufacturing sector has expanded for 18 months and demand remains high.
How Supply Chain Mapping Mitigates Disruption (EPS News), by Bindiya Vakil.
While historically it’s been costly for companies to develop and maintain an accurate map of their supply chain, today, with the right partners, the process can be much more streamlined and efficient.
2021 And Beyond: Preparing for the Next Disruption (EPS News), by Lynn Torrel.
Current realities are triggering a collective reckoning to reengineer supply chains end-to-end and ensure resilience for the long haul.
Resilience—A Moving Target (EPS News), By Vishal Patel.
2021 was a stark reminder of just how closely the world is connected. It further confirmed that the more businesses operate and think like individual entities the more exposed they are.
FAANGs Reshape the Supply Chain (Electronic Products), By Bruce Rayner.
The largest, global, cutting-edge, cloud-centric companies have the wherewithal to design highly sophisticated customized chips, ASICs, and SOCs that set the bar for other lesser companies to try to match.
Want Supply Chain Resilience? Share More (EPS News), by Barbara Jorgensen.
Proposed solutions to supply risks are antithetical to current practices, but a resilient supply chain will require a seamless flow of data between partners.
Ryder: Driving the Digital Transformation for Supply Chains (EPS News), by Barbara Jorgensen.
Ryder has developed proprietary technology to produce business intelligence and analytics. This allows Ryder to provide customers visibility to proactively manage exceptions in the movement of goods through the supply chain. It also connects multiple people to collaborate in real-time to solve in minutes what previously took days or weeks to fix.
Advertisement
