IBM: The Elder Statesman Of Semiconductors
May 17, 2021 Timothy Prickett Morgan
Big Blue may be a lot smaller than it used to be, and it is going to get a lot smaller after its spins off its managed services and hosting business into a new company to be called Kyndryl, which will have about $19 billion in sales, about 4,600 customers, and about 90,000 employees. That will leave the remaining part of IBM, focused on its Power Systems and System z platforms and its whole push into hybrid cloud, artificial intelligence, and quantum computing.
That remaining business will have about $59 billion in sales, about 260,000 employees, and something on the order of several hundred thousand key customers, who buy directly from Big Blue or indirectly through the channel. With the System x business sold off six years ago, that remaining IBM customer base shrank quite a bit but it is hard to say that the System x sale to Lenovo made IBM much more profitable than it might otherwise have been.
Part of the reason that IBM is not as profitable (in the short-term accounting meaning of that word) as it could be is because it invests a reasonably large amount of money into research and development, and even as it has sold off businesses, it has maintained that level of investment or even increased it, as you can see in the chart below.
IBM was about a quarter bigger than it is today back in the middle to late 2000s and it is spending as much money in 2020, and looking ahead into 2021, on research and development. This sounds crazy, but if IBM has learned one thing it is that all of that research and development money keeps its patent portfolio swelling and it can make a decent amount of money selling technology to rivals – including, as it turns out, Intel, which has turned to Big Blue for help on future generations of semiconductor manufacturing as part of a partnership agreement that the two companies inked earlier this year. IBM sold off its merchant and captive foundry business to GlobalFoundries many years ago, and etched its Power8 and Power9 chips with GlobalFoundries 22 nanometer and 14 nanometer processes and is working with Samsung on 7 nanometer and 5 nanometer processes, and as it turns out, is working with Intel and Samsung on creating usable 2 nanometer processes.
Ahead of the Think 2021 virtual conference last week, Arvind Krishna, the chief executive officer at IBM who has been at the helm for the past year, talked a bit about the shortages in the chip business, which are coming at a time when it is getting progressively harder to make each transistor shrink and even harder to make each shrink drive the cost of the transistors down and thus drive the economics of the entire IT industry.
With so many things that it could do with the money it brings in, you might be wondering why IBM is still investing in primary semiconductor design and manufacturing technologies. Wasn’t the whole point of selling off IBM Microelectronics – well, actually paying off GlobalFoundries to take it over is more accurate – to get out of paying for chip fabs as well as R&D? It’s probably a good thing IBM kept doing the research because Intel and Samsung, the latter of which is slated to etch IBM’s Power10 chips and z16 chips very likely its Power11 and Power12 chips and z17 and z18 chips as well, are going to need all the help they can get to develop their 5 nanometer, 3 nanometer, and 2 nanometer nodes.
I went into the details of this 2 nanometer test chip that IBM has created over at The Next Platform, and I am not going to repeat all of the data here. From the IBM i perspective, here is what you need to know. IBM has come up with a technique called nanosheets to stack up transistors much as flash memory cells are stacked. There are three layers of transistors in the 2 nanometer test chip, which is called a 2 nanometer technique not because its transistor gate size is actually 2 nanometers but rather that this 3D technique gives the results in terms of transistor speed and density and power consumption that you might expect from a 2D planar design if it were etching transistors with 2 nanometer gates. (There is a bit of chicanery going on in the chip business, which makes it seem like we are closer to the Moore’s Law limits than perhaps we are. Engineers and scientists bemoan the inaccuracy but also love the drama. Go figure.)
What matters to Power Systems customers is that IBM is partnering with Samsung and has already delivered 7 nanometer, 5 nanometer, and now 2 nanometer test chips in its Albany, New York, tech center on standard 300 millimeter silicon wafers. Those so-called 2 nanometer transistors have transistors that are 45 percent faster or use 75 percent less power at the same speed as the transistors, etched using extreme ultraviolet (EUV) lithography for many steps, that IBM and Samsung have created for the 7 nanometer process being used for the Power10 chips coming out later this year in big iron NUMA machines and early next year for entry machines. While IBM is expecting that 2 nanometer techniques will be in production by the end of 2024, it is not necessarily going to be used on Power11, which we expect to be on a mature 5 nanometer process. IBM is going to keep some of its transistor powder dry and probably use the 2 nanometer technique with Samsung to make Power12 and z18, if we had to guess. (And we do because Big Blue ain’t saying.)
The lack of advanced foundry capacity is a big issue, and one that IBM can help with – for a fee, we presume, of course. And Krishna was asked about the shortages in the chip business, which are hitting all kinds of supply chains these days as more and more customers are chasing fewer and fewer foundries.
“I think that the current semiconductor shortage is very worrisome,” explained Krishna. “While there has been a lot of attention on the shortage of chips for automobiles, and that’s an important problem, it goes much further than that. Is there a shortage of just those, or is there also a shortage of chips for memory? And that impacts that laptops and tablets for students in K-12? Or is there a shortage of chips for servers? And in consumer electronics, it’s across the board. It is no longer just one of those. People are being asked to make what I call a really unfortunate dilemma, which is what should get prioritized? And that’s really hard to do. Now, if we want to increase capacity, especially closer to any particular geography, then the time scale to do semiconductor increases is measured in years, not measured in months, unfortunately. It is not so simple to take a design that was done for one particular manufacturer, for one particular plant, and just take it and drop it somewhere else.”
IBM should know, because the problems that GlobalFoundries had with 7 nanometer EUV processes forced it to shut down that manufacturing, pushing IBM into the arms of Samsung, the only other choice besides and overloaded Taiwan Semiconductor Manufacturing Corp, which is making most of the advanced devices on the planet.
Looking ahead, IBM is clearly positioning itself to help Intel and its Intel Foundry Services merchant foundry business, and we highly suspect that Big Blue will want an indigenous second source for chip manufacturing so it can ensure that Power11 and Power12 chips, as well as z17 and z18 chips, come to market more or less as planned before the end of this decade. And that is about as far as anyone can see in the IT business.