Update: CadenceLIVE India, Ransomware, 2nm, and More

This is another one of my update posts, following up on topics I covered in earlier Breakfast Bytes posts but where the new material is not enough to justify an entire new post by itself

CadenceLIVE India

CadenceLIVE India, originally scheduled for August 4 and 5, has been postponed. A new date has not yet been announced.

I always write about the season of CadenceLIVE events early in the year, but since that is now out-of-date, it seems silly to link to it. Instead, the most accurate information is always available on the CadenceLIVE event page.

Colonial Pipeline

This is a followup to my post Evolving Maturity in Ransomware. Among other things, this post pointed out how much worse ransomware has got over the last eighteen months or so, and pointing out it was likely to get worse. Well, less than a month later it has. I live in California, so I'm not directly affected, but the Colonial Pipeline, which I'm sure few people had heard of until a week ago, was shut down by a ransomware attack. This pipeline carries a large percentage of the fuel used on the east coast. In no time, there were lines at gas stations. Worse is when the trucking industry cannot get fuel since then the shortages spillover from gasoline to food. In addition, the hackers stole 100 gigabytes of data and are threatening to publish it. The White House declared a state of emergency and created a task force to deal with the problem, but it is unclear what they can do.

As I write this at the end of last week, there are reports that a ransom of about $5M was paid. See, for example, the Wall Street Journal piece Colonial Pipeline Said to Have Paid Ransom to Hackers Who Caused Shutdown.

That wasn't the only recent ransomware attack. Scripps Health in San Diego was also partially shut down by a ransomware attack. See Ransomware Hits Scripps Health, Disrupting Critical Care, Online Portal.

And in other security news, but not ransomware, vulnerabilities in Wi-Fi were announced. See FragAttack for technical details. One surprising thing is that some of the vulnerabilities are in the Wi-Fi protocol itself, not in poor implementations. Some are so fundamental to the Wi-Fi protocol that they go back to when Wi-Fi was first invented and have always been there. It reminds me a bit of Spectre and Meltdown, security issues with out-of-order processors that have been hiding in plain sight for 20 years.

I talked to a couple of security experts to ask about the technical details of the pipeline security breach. The details are not known, but this is probably very mundane: default passwords, unpatched systems, that sort of thing. In some ways, this makes it scarier. It is not a sophisticated attack like SolarWinds. More like the NHS attack a few years ago which stemmed from using unpatched Windows-XP systems after it had got so old that even Microsoft no longer supported that version of the operating system.

Increasingly, those of us that have anything to do with chip design are in the firing line for improving security. Software-only solutions do not work, nor do solutions that depend on companies that are experts on other things (such as running pipelines or hospitals) becoming security experts. We have to build it into silicon in a way that incorporates best-in-class solutions. Plus, we need to take especially good care of our own supply chains, or who knows what we might be building into chips. The Solarwinds breach is a shot across our bows.

IBM Announces 2nm Process

In some ways, this is a followup post to any number of posts about process roadmaps at semiconductor manufacturers. But it is also a follow up to one of the first posts I ever wrote on Breakfast Bytes less than a month after I rejoined Cadence: Where Does 5 Really Mean 30? Process Node Naming. That pointed out that since there is nothing that is 5nm on a 5nm process, the name of the process is pure marketing.

So is IBM's process really 2nm?

A bit of history. IBM used to be one of the leaders in semiconductor manufacturing, using its own chips in a wide range of mainframe products. Then, it entered the ASIC business, which I remember since it used Compass libraries. In 2014, it "sold" its semiconductor manufacturing business to GLOBALFOUNDRIES. I say "sold" in quotes because it actually paid GF to take it off its hands. However, it kept a portion of its semiconductor research organization. The expectation is that GF would work its way down the process node roadmap and supply IBM with wafers. But that turned out to be too big of an investment for GF who pivoted in 2018 and decided that 12nm would be its most advanced node. It sent back its EUV stepper to ASML and focused on specialty processes, especially FD-SOI (which they called 22FDX and 12FDX). That left IBM high and dry since it still needed advanced-node chips for mainframes. In 2018, it announced that it was partnering with Samsung. Samsung has already announced a 3nm process that will use horizontal nanosheet technology (also known as gate-all-around or GAA).

The 2nm that IBM announced is also a horizontal nanosheet technology. Here is the press release. The above picture is from the release. If IBM was hoping for headlines, it certainly got them, with breathless pieces written about how IBM had leapfrogged everyone. My spidey senses went off when the press release said "50 billion transistors in a 'thumbnail'-sized area".

My go-to guy for all things to do with semiconductor manufacturing is Scotten Jones who runs ICKnowlege and makes a business selling cost models so manufacturers can cost out process decisions and also evaluate the likely costs of their competitors. He was a colleague at Semiwiki before I rejoined Cadence, and still writes there. In a piece titled Is IBM’s 2nm Announcement Actually a 2nm Node? he concludes that by the normal standards of foundry naming, that it is not.

Here are the details that IBM announced, which Scotten uses as his starting point:

• "2nm”
• 50 billion transistors in a “thumbnail”-sized area, later disclosed to be 150mm2 = 333 million transistors per millimeter (MTx/mm2)
• 44nm Contacted Poly Pitch (CPP) with 12nm gate length
• Gate All Around (GAA), there are several ways to do GAA, based on the cross-sections IBM is using horizontal nanosheets (HNS)
• The HNS stack is built over an oxide layer
• 45% higher performance or 75% lower power versus the most advanced 7nm chips
• EUV patterning is used in the front end and allows the HNS sheet width to be varied between 15nm to 70nm, which is very useful to tune various areas of the circuit for low power or high performance and also for SRAM cells
• The sheets are 5nm thick and stacked three high

Final conclusion, it should be called a 2.9nm process by foundry naming standards (or a bigger number by Intel naming standards):

Lots more detail in the deep dive Scotten makes at his post on SemiWiki. Since IBM no longer has semiconductor volume manufacturing, it is unclear how this technology will be made available, if at all.

Arm Neoverse N2 and V1

I wrote about these new data center processors and the Cadence Rapid Adoption Kits (RAKs) in my post Rapid Adoption of the Arm Server-Class Processors. Since then Arm has produced a video tutorial on the new processors.

China Lands Rover on Mars

I wrote about the three spaceflights to Mars that left the US in July 2020 in my post It's Mars Month. Of course, the US successfully landed the Perseverance rover and even has successfully flown its helicopter Ingenuity. Last week, China successfully landed its Zhurong rover. As the BBC puts it:

The vehicle used a combination of a protective capsule, a parachute, and a rocket platform to make the descent. The successful touchdown is a remarkable achievement, given the difficult nature of the task. Only the Americans have really mastered landing on Mars until now. All other countries that have tried have either crashed or lost contact soon after reaching the surface.

Here's an animation of the landing:

The Biggest Solar Storm of the 20th Century

The biggest solar storm of the 20th century took place 100 years ago last weekend, on May 13-15th 1921. As it happens, I just discovered last week that my post The Carrington Event: When Will We Have Another? is the second-highest viewed post ever on Breakfast Bytes. Recent speculation is that this solar storm was as big as the Carrington event, although it is not really known since there was no accurate measuring equipment in the 19th and early 20th centuries.

The summary paragraph from Wikipedia says:

The three-day May 1921 geomagnetic storm was caused by the impact of an extraordinarily powerful coronal mass ejection on Earth's magnetosphere. It occurred on 13–15 May as part of solar cycle 15, and was the most intense geomagnetic storm of the 20th century. Since it occurred before the extensive interconnectivity of electrical systems and the general electrical dependence of infrastructure in the developed world, its effect was restricted; however, its ground currents were up to an order of magnitude greater than those of the March 1989 geomagnetic storm which interrupted electrical service to large parts of northeastern North America. The storm's electrical current sparked a number of fires worldwide, including one near Grand Central Terminal which made it known as the "New York Railroad Storm".

Sign up for Sunday Brunch, the weekly Breakfast Bytes email