COVID: The Complete Report
June 26, 2020 — Strategy and Tactics: COVID: The Complete Report WildPhotons: If you don't change directions you'll ...
We've had so many requests to give broader distribution to this report on the Coronavirus that we are releasing it for anyone to read. It was the result of recent arguments with family members who had picked up really wrong ideas on Coronavirus data and the collection of it from social media sites that seek to dismiss the danger of this disease and frame as a political and personal rights issue. Now I did not want to get into the political issue, because this is personal. But I did want to fact check the points being made to see if they were true and arm myself for next week's American Independence Holiday weekend. So, I did what I do with data. Anyway, I thought I would share them with you, as I am sure you've run into others who have been swayed by these social media sites. I know it's not normal Chip Insider material, but it will arm you for discussions that may come up around the barbeque grill next weekend.
"Intelligence is quickness in seeing things as they are"
— George Santayana
VLSI 2020 Symposium:
Normally held in Hawaii, VLSI 2020 showed how participation could zoom up when the travel barrier is removed — plus lowering prices. There was an 18% increase in registrations. The IEEE also had a record 248 submissions to the Technology side of the conference — more than 30% greater than in the previous three years. Of the accepted papers, Beyond CMOS shrunk, while "Device Technology for New Computing" grew by almost three times. Memory was up 20%. What was most interesting was the new bundle of Device Physics, Characterization, Modeling, and Non-Si Substates/Materials Devices went from nothing to bigger than memory in a single year. It illustrates the shift of semiconductor R&D from 'little' r&D to 'big R&d.' This was further confirmed on the Circuits side, where the declining trend in the upper channel since 2014 continued. The growth in AI, Security, and Machine Learning was nil, compared to previous years. This indicates the research action is shifting from circuits to technology. And the technology focus is on finding ways to improve PPAC (Power, Performance, Area, and Cost).
Once in the conference some interesting trends could be seen. First, and least surprising, was that work in nanoribbon GAAs as a replacement for tri-gate FinFETs was hot. Kangguo Cheng of IBM presented an improved air-space integration scheme that is "agnostic" to transistors below (THL.5). While this is an old technology, it's difficulties with integration have always limited its use. One of the big issues for the industry has been that the k in low-k dielectric films has been stuck for a decade or so. It's important because it limits device speed, and is especially important because copper is also offering limited improvements. IBM showed a 15% improvement in Ceff (effective Capacitance), which is significant (for more on this check out this impact appraisal, which I wrote in 1995. It's old but still relevant).
The Last Wavelength: An emergent trend I saw was researchers appear to be internalizing the fact that EUV is the last wavelength and pivoting their work to do something about it. One indication is the rapid revival of DSA (Directed Self-Assembly) since 2019, when researchers would tell you it "had missed its window." Earlier this year, at SPIE, I noticed DSA was garnering more attention than usual. But it was at VLSI 2020 that I could see an emergent trend. Now some might be led to believe that it's now the heir-apparent to EUV. However, lithography technologies seldom die. Instead they fade from the critical, to the less critical, to the not so critical in what is a decades-long process. Still, I doubt this will even be the case because being the last wavelength implies all efforts will shift to optimizing and extending it like what happened with 193i and George Gomba predicted would happen with EUV two years ago. It also means the opportunities will shift back to materials-system-engineering, where companies like Applied Materials have done so well.
One of the best talks was the plenary from Mike Mayberry of Intel. What makes for a great plenary, is the ability of an executive to bring together disparate concepts and put them together in new ways to cause the audience to think differently about the road ahead. And Mike delivered by laying out a new Cost-of-Ownership problem for our industry: data transport.
The original big CoO problem was laid out by Gordon Moore in 1965 and we call it Moore's Law. The big CoO problem was how to lower the cost of compute with the solution of scaling. The next big CoO problem came around 1988, when the term was invented by Sematech. The big CoO problem was advances in technology were no longer outpacing increases in the cost of the technology. The solution came in the form of 300mm, higher tool availability, and throughput increases that would lead to the bone-crushing G-forces in today's scanner stages. Now the importance of this walk down silicon history lane is to portray the idea that big CoO problems create opportunities for everyone that can solve them and that they drive growth for decades.
Data Transport's CoO problem — as Mike laid it out — is that the cost of moving data exceeds the computational cost of creating it. This is compounded by the 26% annual growth in data being made possible by the growth in device production. Mike sees the solution as starting with a shift from the centralized data center topology where the data is brought to the computational engine dominating the last two decades. That shift is to a distributed topology, where the compute is brought to the data. It is much like the era when compute went from the mainframe to the mini-computer, to the PC. But now it is from the hyper-scale data center, to the regional data center, to a cloudified data center, to a micro-data center, and on to "intelligent agents" (beyond IoT: phones, watches, cars, security cameras, medical imaging and more). At the chip level, Mike sees this driving domain specific architectures, such as Amazon's Gravitron, Google's TPUs, and others from Alibaba, Baidu, Huawei, and Adobe. He used the 127mm2 A12X APU as an example of what's happening at the die level: with 4 big CPU cores that take up 8% of the die area and a 7 core GPU that consumes 24% of it. But the fastest growing part of the die is the NPU and only 4%, yet is 4x compared to the previous generation. Now if you think back, this same thing happened a while back with the GPU, which was taking over die area from the CPU. Yet die area grew, so it's good news for silicon demand, with more Reticle-Limited-Designs (RDLs) on the way.
Now if you know Mike, you know he never gives one-trick-pony presentations. After that, he drilled down into why Gate-All-Around is displacing Tri-Gate, which should be obvious with the latter's less common name for finFET. What's not so obvious is that GAA, with nanosheets of 2-D materials are scaled vertically, not horizontally as finFETs are. Then he discussed DSA with a slide fittingly named, "Better Computing through Chemistry," showing how it can tighten up spacing variation. And then went on to discuss monolithic 3D transistor stacking, 3D heterogeneous integration with chiplets (the solution to RDLs), and much more.
Then he took us through Neuromorphic computing — did you know a Ladybug has 250-500K neurons, a Cockroach 1M, and a Zebrafish 10M; the point of which these compare to successive generations of Intel releases from 2017 through 2019 — which he is much more bullish about. And quantum computing, which he believes is a decade away from practical use. Finishing with Beyond CMOS device architectures, he described the Magneto-Electric Spin-Orbit cell they had working in the lab today.
As he finished, I began to wonder if the metric of chasing nodes was dropping in importance. That the future will be some melding of nodes and architecture in a new metric yet to be conceived. He also took the prize for the most interesting video call background: A Police Box screen that captivated my attention. Thanks Mike!
As the conference came to an end, I realized it was one of the best in terms of indicating clear strategic inflection points: both technology-wise and in terms of how conferences are presented. If one thing is clear, we probably won't go back to the before times. That said, you can't replace the interaction of face-to-face meetings. While heathier, there is a cost to replacing jet-lag with the mental stress of a day of 2D video calls, where your eyes and brain strain to gain the 3D cues that are not there on a flat screen — even a curved one for that matter.
What worked well in this virtual conference: It was a lot easier to fit it in than I thought it would be. Plus having the ability to rewind and catch things I missed. The e-mail stream they gave was a great reminder as to what had been missed by the normal daily interruptions that had taken me away. These were benefits no live conference could have.
What did not work so well in this virtual conference: All of the normal random interactions that happen in the halls of a conference were sadly missing. Often this is where the best information is found. Also missed was the head-clearing separation from the office that a conference brings. That said the Q&A discussions were good, often because they were separated from the presentations. The bottom line is this has changed everything. Even if a physical conference comes back, I believe the organizations that can adapt best to the new normal by blending content for the virtual conference and a physical conference will have the competitive advantage going forward.
Coronavirus Watch: I recently found myself in some arguments with family members who had picked up really wrong ideas on Coronavirus data and the collection of it from social media sites that seek to dismiss the danger of this disease and frame as a political and personal rights issue. Now I did not want to get into the political issue, because this is personal. But I did want to fact check the points being made to see if they were true. The most egregious of which, is a belief that all deaths are being classed as caused by COVID-19. My dad has actually been collecting the daily Coronavirus data into a spreadsheet and passing it along to me, which combined with more data from other sources, I did what I do with data. Out of this came some charts to make my case. The sources are from the WHO, CDC, NHTSA, NSC, Johns Hopkins University, The California Department of Health, and the Bay Area News Group. Anyway, I thought I would share them with you, as I am sure you've run into others who have been swayed by these social media sites. I know it's not normal Chip Insider material, but it will arm you for discussions that may come up over next week's American Independence Holiday weekend. So here goes:
First off is the belief that that all deaths are being classed as caused by COVID-19. The data from the CDC depicts a very different story, where the total of all the deaths they report is 10X the Coronavirus cases they report.
Another is a belief that COVID is no more dangerous than the flu. The CDC collects extensive data on the flu every year, so here I have used the entire count of deaths seen in the United States over 2018-2019 influenza season. Since Coronavirus infections have not subsided, the 34K who died from them makes for a very conservative comparison. There are better ways to look at it, using statistical metrics health professionals use.
Epidemiologists have three key measures of the severity of a disease, which many people discussing it on social media appear to be confusing or ignoring. These are the:
Plotting the infection rate sums of the COVID since March and through June 24th yields a steady growth. What's most interesting about it is it dispels the idea that Shelter-in-Place, or SIP does not limit infection rates, because Santa Clara County and California have much lower infection rates than the USA as a whole. And I don't need to mention Santa Clara County and California were arguably the most aggressive to implement SIP and mandatory mask wear. In broader words, politics is an opinion about reality, science is a factual reality.
- Infection rate: The ratio total people infected divided into the total population
- Mortality rate: The ratio total people who die from the disease divided into the total population
- Fatality rate: The ratio total people who die from the disease divided into total people infectedn
Mortality rates, or the ratio of total people who die from the disease divided into the total population, also show a graduated slope for COVID-19. Along with the same demonstration of the value of SIP, Santa Clara County is the only region where you have a lower chance of dying from it than the flu or from being in a car accident.
Now remember, the mortality rates only consider the probability you might die from it. Fatality rates add the condition that you have already been infected.
Fatality rates show the first positive news, where the curves have peaked and are declining. And this is even true on a worldwide scale. One might conclude several perspectives from this such as: the severity of the virus is attenuating or the health system is improving. Both of these causes are possibly true, but the degree to which is true has yet to be proven.
It also contradicts some opinions of political leaders that more testing makes it look worse, because more testing should give medical professionals the ability to treat people earlier and thus lower the fatality rate. The fatality rate is clearly not increasing due to more testing because the downward slope correlates to the greater availability of tests kits. Another opinion that can't be demonstrated is that COVID-19 is no more dangerous than the flu.
It is far more dangerous than the flu or even car accidents. While the fatality rate is declining, you are still 54 times more likely to die from the Coronavirus if you get it than the flu. And 8 times more likely than if you are in a car accident.
The other opinion floating around out there is that it is not dangerous for younger people so that means we should allow them the freedom to run around. Let's start with the data, which shows — as you might expect — that fatalities lean far more to the right of the age scale. While that's true, I wondered if it was a false conclusion because this should be true for the population as a whole. After all, I'd just been to a graveyard to see the latter. But what about the overall data? It does demonstrate age is a function for most disease-related causes of death. Deaths in the 0-4 and 15-16 years old classes are nil, while 65 and older account for 74% or more of the total.
A different perception emerges when each category is normalized to the flu:
Which brings us back to the political question of should we let younger children run around — or more poignantly send them back to school. This has a three-way punch of scientific, moral, and political and aspects. The scientific is a question of transmissibility. Unfortunately, it is hard to put metrics on COVID-19 transmissibility because it is so new and there are so many variables affecting transmission. One measure I see is the fast rise of infection and mortality rates, which has only been slowed my personal distancing, contact tracing, masks, and quarantines.
The moral question is one I remember well from an ethics class I took in my University's Philosophy department. The moral question has no clear boundary line between the "unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness" — as stated in the Declaration of Independence— and to what degree individual rights can infringe on the rights of others of Life, Liberty and the pursuit of Happiness. In the case of COVID-19, life is clearly infringed upon by transmission — Liberty is if you are confined — And Happiness if you are dying or dead. In this case, it's a question of the rights of one class to kill others by transmitting a disease. On this moral point: In the case of AIDS, courts have ruled if you know you have it and infect another... it can be treated as murder. This is where the issue shifts to the political, or the degree to which the state can put restrictions on individuals to protect the health of the whole. I'm going to stop here, because there is no data driven decision to be made. Keep this on your phone, as it may help you in the arguments around the barbeque grill next weekend. Just be sure to have a beer in your other hand.