The trajectory of the pandemic in 2021 may or not influence technology trends. Even so, I think a few trends are worth mentioning...
The end of the year is often a time of celebration but almost always one of emotion. From resolutions for the year about to launch to regrets on a year that has past, we mostly look to end the current year quickly. The demise of 2020 is causing the scribbling down of my thoughts on the year ahead more rushed.
For many years, industry trends have focused on mobility and semiconductor technology has largely served those trends. Investment and developments in cloud computing have grabbed a lot of the spotlight in the last few years, but much of that also addresses mobility. It was the technology mantra.
Then came 2020, a year that many would gladly forget. But not the information technology industry. It did quite well. As work shifted home, more data shifted onto the cloud and more remote access was necessary. But frequent waves of infections and subsequent lockdown orders around the world reduced the utility of mobility. Yes, workers were not centralized, but neither were they editing PowerPoints and emailing the boss at happy hour.
I am no more qualified to comment on grand theories about the return to work any more so than the epidemiology of SARS-CoV-2. (Alas, if it were only qualified folks who voiced opinions on epidemiology. Perhaps a few New Year’s resolutions are in order. I will not provide any, but they are welcome in the comments.) The trajectory of the pandemic in 2021 may or not influence technology trends. Even so, I think a few.
Snapdragon 888 – Coming very soon to a phone near you
The new Snapdragon 888 will be the third major product launched on TSMC’s 5 nm node, the most advanced mass production process technology available (Apple’s A14 and M1 hit the market first). It just so happens to also be the second mobile application processor, but the Snapdragon 888 exceeds the competition in several key features.
Three primary areas of improvement for the 888 are camera capabilities, gaming performance, and AI. As Jim McGregor pointed out in his coverage of the Snapdragon Tech Summit, the new Snapdragon “will feature the X60 RF modem (modem + RF solution); an enhanced 6th generation AI engine with a new Hexagon processor, a new sensing hub, and 26 TOPS of overall performance; and a new Adreno GPU with more performance over the previous generation.”
The camera functionalities are a little over the top, so this was detailed in McGregor’s second piece on photography and gaming. The Snapdragon 888 will deploy three image sensor processors that can blaze through 2.7 gigapixels per second. Serious photographers may lament this as something of a pixel race, but for regular folks, the 888 allows three individual image sensors to each acquire 4K stills or video in 10-bit HDR simultaneously. Not bad.
The main milestone from an on-chip functionality perspective is the inclusion for the first time of a full 5G modem in a system-on-chip (SoC) since the Snapdragon 888 includes the X60 5G modem. The Apple A14 did not include this. Instead, the iPhone 12 family used a separately packaged Qualcomm X55 modem and Qualcomm RF components like the SDR865 transceiver.
Despite the rise of Mediatek and the the Exynos 1080 that Samsung will soon hand the flag to, Qualcomm appears to be holding its lead in the mobile space in large part due to its prowess in RF technology. Recent reports that have Mediatek in the worldwide lead for mobile chipsets still place Qualcomm as the number one for 5G. The early days of 2021 will provide more details as Mediatek, Qualcomm, and Samsung catch up to Apple with flagship application processors in 5 nm.
Extreme performance aside, the first use case for the most advanced mobile application processor every produced will be better cat pictures from the couch more likely than not uploaded through an obsolete home WiFi network. Despite all that, the Snapdragon 888 (or ba-ba-ba as some refer to the Chinese lucky numbers) represents the pinnacle of SoC development.
Qualcomm notes a few key selling points. “Featuring our completely rearchitected 6th gen Qualcomm Artificial Intelligence (AI) Engine, the Snapdragon 888 5G delivers a total of 26 TOPS performance, 3 times performance-per-watt improvement and 16 times larger shared AI memory.”
The newest Snapdragon is an octa-core design with one high-performance Cortex-X1 core, three Cortex-A78 cores, and four low-power Cortex-A55 cores, an Adreno 660, three ISPs, and the 6th generation AI engine all point to a healthy SoC market and continued development. That’s a lot of cutting edge design, but there a couple of things to draw out. The Snapdragon 888 is the first chip to include the Cortex-X1 core. It is worth noting the “16 times larger shared AI memory” since we may expect to see more chip real estate devoted to SRAM cache than the captive design of Apple’s A14.
Several handset brands have announced deployment of the Snapdragon 888 and some may arrive before the ball drops on Times Square. The 888 will be the hottest SoC story of 2021 by several measures, but it may not be the only one. In one of the major milestones of 2020, Apple began deploying its own ARM-based SoC designs to its personal computer line. Indications are beginning to point to Microsoft following suit.
A hot topic of the last year (and sometimes hotly debated as it was in the comments to Brian Santo’s conversation with Ramune Nagisetty of Intel) was the shift away from system-on-chip design and toward a system-in-package approach using chiplets. How apropos for this trend to take hold during the times of “physical distancing” as technology imitates life in the pandemic by pulling apart the IP blocks of monolithic silicon integrated circuits and spacing them out in multiple chiplets assembled onto package substrates.
The idea of physically separating IP into pieces of silicon rather than stitching them together monolithically on a single die has engendered many names from “chiplet” to a range of other labels like the tried and true system-in-package (SiP) or the trendier heterogeneous integration technology (HIT). The various names have all attracted a lot of attention, especially my own. This new trend (I will risk negative comments by calling it “new” despite the multi-chip module, or MCM, and other SiP variant technology and labels being anything but new) has likewise attracted defense funding to nudge along a new IP ecosystem to allow cutting edge technology outside the realm of the traditional SoC.
Since monolithic SoC design does not fit well with Department of Defense (DoD) or other low-volume applications, DARPA launched and provided funding for the Common Heterogeneous Integration and Intellectual Property (IP) Reuse Strategies (CHIPS) program seeks to establish a new paradigm in IP reuse.
From a structure and materials perspective, there are many options already available with proven results. From silicon interposer 2.5D designs with high bandwidth memory for high performance computing and GPU to cost-effective fan-out like TSMC’s integrated fan-out (InFO) wafer level packaging, there are options currently available to address a wide range of product applications.
To create a new ecosystem for chiplets, however, more work especially in standardization is needed. This work is not likely to be completed in the next year, but expect major strides in that direction. The basic commercial economics are there to drive players outside the SoC club of the Apples and Qualcomms, not to mention the need for a viable option for military hardware to stay at the cutting edge.
In order to create a practical ecosystem for non-SoC players to take full advantage of the chiplet approach, some standardization of inter-chiplet communication is necessary. A few approaches have been touted. It will take some time to evolve, but the coming year may give more insight into which approaches may emerge. There are some proprietary interconnect schemes floating around, but the key to this approach is going to be interoperability of chipsets and streamlined integration of each piece of IP into the broadest possible set of product use cases. In other words, there needs to be a market for chiplet vendors.
Synposys suggests one high speed serial interconnect option: “As a result, high-speed die-to-die (D2D) communication is needed to pass large data sets between die within a chip [sic]. Ultra Short Reach/Extra Short Reach (USR/XSR) SerDes make this possible, with current designs using 112Gbps SerDes and higher speeds likely to come within the next couple of years.”
Intel has the advanced interface bus (AIB) that it has made available with a royalty-free license since 2019. The AIB specification indicates 2GB/s/wire with channels of 40 wires currently used and supporting up to 160 wires per channel. The AIB standard is created with Intel’s Embedded Multi-die Interconnect Bridge (EMIB) in mind. The first generation of AIB was deployed in the Intel Stratix 10 products. Intel promises that AIB has lower latency than SERDES approaches making it more suited to a wider array of chiplet types for heterogeneous integration.
There are at least a few more interconnect options. The Open Domain-Specific Architecture (ODSA) group is working on two die-to-die interfaces—Bunch of Wires (BoW) and OpenHBI. The point being there is a lot of diversity at the moment. As a consensus forms on interconnect schemes, the viability of chiplets in the marketplace will accelerate. I am getting behind BoW, just for the name.
A Post-FinFET Future
In typical fashion, the finFET age of semiconductor process technology has persisted well past original roadmap predictions. The concept of stretching transistor channels into the third dimension to improve gate electrostatics and control conduction in the channel first came to market under the Intel moniker, Tri-Gate. Manufacturers have kept the finFET viable through the 5nm node with, among other innovations, the use of high mobility channels which use alternatives to pure silicon.
Although some see the end coming sooner than others, the end is coming nonetheless. With the 5 nm process in production in 2020, Apple’s processors were still using TSMC finFETs. And TSMC is one manufacturer planning to squeeze another full generation out of finFETs as they have announced them for their 3 nm node.
The replacement for finFET will come from a technology in the “nanowire” or gate all ROUND (GAA) category. Early technologies announced for this approach are actually a flat shaped wire or “nanosheet.” Samsung has announced a shift to what they have branded the Multi-Bridge Channel FET, or MBCFET, (MBC) for the 3 nm node. Some may say promising this node in 2021 is aspirational. It is certainly aggressive, but for me it is the most exciting technology news to watch for in the coming year.
Tech Cold War
The U.S. government appears to understand the strategic importance of the semiconductor industry. Congress has proposed a bill called Creating Helpful Incentives to Produce Semiconductors for America Act, or CHIPS for America Act. Not to be outdone, the U.S. Department of Defense through its Defense Advanced Research Projects Agency (DARPA) has created the Common Heterogeneous Integration and Intellectual Property (IP) Reuse Strategies (CHIPS) program to drive the chiplet ecosystem.
U.S. government initiatives received well-deserved media coverage in 2020.
The year was also the setting for an ongoing US-China trade war with the semiconductor industry an integral piece. Junko Yoshida’s thoughts from several months ago shed light on the state of the conflict with China. EETimes created a special project devoted to this topic that is worth reviewing: The New Tech Cold War.
There is no doubt that the chip business is strategic and the U.S. government gets it. Among the steps taken was blocking U.S. chip companies from supplying Chinese equipment makers, chiefly Huawei. Domestic IC manufacturing has struggled, particularly for advanced nodes, and China depends upon foreign production. Although some may see development of viable Chinese semiconductor production as an inevitability, the U.S. administration has chosen to deprive China of the production tools needed to build cutting edge fabs.
With China fully dependent on Taiwan for state-of-the-art processes for consumer devices and telecom equipment alike, restrictions were placed on TSMC for delivery to Chinese fabless companies like HiSilicon.
History and the current trade posturing leave Taiwan in an unenviable position in the shadow of one superpower and closely allied with its nemesis. How hungry that big tiger might get looking down on Taiwan opens the door to a lot of speculation, some of it difficult to ponder.
A recent NY Times op-ed laid it all out right there in the title, “Pound for Pound, Taiwan Is the Most Important Place in the World.” Ruchir Sharma gives a brief history of Taiwan with emphasis on the unprecedented rise of TSMC.
David Pierson and Michelle Yun go into more detail about the trade disputes in the LA Times. Their article title is meant for a wider audience and will draw a few smirks from EETimes readers. But “The most important company you’ve never heard of is being dragged into the U.S.-China rivalry” is still worth a look.
With the U.S. set to welcome a new President in January, I wonder what a new administration will mean for the tech cold war. Many expect the Biden White House to be a radical departure from the previous four years, but will it engender much change in approach to China?
Abishur Prakash had an interesting way of looking at the question as he was quoted on CNBC. “The bullet has left the chamber.” During the Trump White House years, sanctions and executive orders were used extensively to challenge China. The status quo in Chinese relations will likely be unaffected by the more general mood change set for the White House. Use whatever metaphor you like, but Elvis has probably left the building.
One final thought on the tech cold war: Perhaps a better way to think about it is, “Why worry?” In Forbes, George Calhoun describes why the U.S. will remain dominant in the semiconductor business and the daunting task China faces to catch up.
A major milestone in the tech cold war was the announcement that TSMC will build a fab in Arizona. Some have questioned the feasibility, but I think we will see continued progress toward this goal in 2021, slow and steady though it might be.
If procrastination was not enough for me to overcome to finish this year ender essay, news seems to be stacking up on New Year’s Eve. Another trend to watch for this year is the long term strategic direction Intel will follow. Reuters is just now reporting that an activist hedge fund has begun to push the Intel board to consider alternatives to remaining a fully integrated device manufacturer. Intel may not take significant strides in the footsteps of AMD and begin to spinout the foundry operations, but the speculation is just one more reason to watch Intel closely.
Final Word (I promise)
Let me wrap 2020 with a few quick thoughts (with apologies for self-indulgence). First, I am grateful for the opportunity the past year presented me for restarting my contributions to a first-class media outlet like EETimes.
Second, I want to thank everyone who read the columns. It is always encouraging to know that someone besides the editor has read your stuff.
Third, there were many kind comments left by readers that were a welcome reward. The anonymous internet and license to ridicule that social media has given make thoughtful, positive comments a particular joy.
But I am also grateful for the engagement of all who took the time to comment with a differing opinion or a correction to my own analysis or the facts presented in the articles. One of the great wonders of the EETimes readership is that the corrections and kind comments often came from the same person.
Fourth, thank you, Brian Santo, for your excellent editing work. You improved everything I submitted. Editors have a tough job that requires tirelessness. Brian does all that with a friendly smile. He’s just an all-around cool guy. I’m happy to be working with him.
My final thoughts go to Junko Yoshida. So many know her through her excellent writing, a working relationship, or both, so most of the people reading this will already understand. I would fall short trying to add to the obvious. As a contributor, she always makes my effort seem worthwhile.
All the best for 2021.