Our industry is difficult to understand. Most of us resort to imperfect analogies to explain it when we have to. Some of this is because we have a complex supply chain for creation. It is easy to get confused between an instruction set architecture (ISA, like x86), microprocessor IP (like Arm processors), an SoC containing a microprocessor (any smartphone application processor), a chip that is an entire microprocessor (many Intel products), a fabless semiconductor company (say NVIDIA), a foundry (GlobalFoundries), a fab (like Fab 1 in Dresden), subcontract manufacturing (like Foxconn in Shenzhen). And confusingly, some companies do multiple things: for example, Samsung is a foundry, has fabs, is in the memory business, sells smartphones, TVs, solid-state disks (SDD), and more.

The good news is that there are two articles on our industry in the current issue of The Economist, A new blueprint for microprocessors challenges the industry’s giants about RISC-V, and India is trying to create an indigenous chip-making industry. The bad news is that they show how even professional journalists can get completely confused about the details.

RISC-V

The RISC-V article starts off getting it mostly right:

Most microprocessors—the chips that do the grunt work in computers—are built around designs, known as instruction-set architectures (ISAs), which are owned either by Intel, an American giant, or by Arm, a Japanese one.

I think that is the first time that I've seen Arm described as a Japanese giant, that's on par with calling Guinness an English beer or Budweiser a Belgian one. But Arm is indeed owned by Softbank, which is Japanese (and Diageo is based in London, and InBev in Leuven—yes, there is more in Leuven than imec).

I said above that most of us resort to imperfect analogies to explain our business—for example, in the EDA 101 presentation that I give internally at Cadence from time to time I compare chip design to designing a commercial aircraft, about the only thing that comes close in complexity. But The Economist's analogy for an ISA is...weird:

To draw an analogy, a house might have two floors or three, five bedrooms or six, one bathroom or two. That is up to the architect. An ISA, however, is the equivalent of insisting that the same sorts of electrical sockets and water inlets and outlets be put in the same places in every appropriate room, so that an electrician or a plumber can find them instantly and carry the correct kit to connect to them.

Just going with the electrical socket (outlet) would have been better: any plug can go in the outlet, anyone can design the wiring behind it, if the arrangement of the pins is not proprietary then anyone can design the outlet itself.

Then the article gets really confused:

RISC-V offers computer architects a way to standardize their sockets and plumbing without having to gain permission from (and pay royalties to) either of the monopolists.

I'm not sure that's even true. AMD builds x86 chips, for example. In fact, technically, it's their ISA since 64-bit came along while Intel was off on its Itanium detour. In any case, Intel and AMD are not in the instruction set business, nor even the microprocessor IP business, royalties or no royalties. And Arm is not in the instruction set business either—I don't think you can license just the instruction set without licensing processor (and other) IP.

A second, more subtle advantage is that, unlike chips based on proprietary designs, those involving RISC-V can be used without lengthy and expensive contractual negotiations. It can take between six months and two years to negotiate a licence to use a chip design involving a commercial ISA.

That's because you are not negotiating to get access to the ISA so you can design your own microprocessor. Intel and AMD are not in the processor (or ISA) licensing business, and Arm will ("after 6 months to two years") license you the IP for a processor, not just an ISA. Which, eventually, The Economist notices:

RISC-V does have weaknesses. Arm has spent decades building software tools to work with its designs, and spends a lot of its time helping customers implement these on their chips. The tools that exist for RISC-V designs are not yet that sophisticated. Intel makes things simpler still. It carries out all of the development, testing and fabrication itself, delivering only finished chips to customers. This reliability will certainly keep these firms’ products competitive for a while.

This is another weird comparison since RISC-V is an ISA, and Arm licenses microprocessors not ISAs, and Intel doesn't license anything. And AMD builds chips that compete with "monopolist" (the Economist's word from the start of the article) Intel.

The last paragraph gets even weirder:

Open-source software was a prerequisite for the smartphone boom that has taken place over the past decade. Open-source hardware, such as RISC-V, may lead to a similar expansion of other devices in the decade to come.

In what way was open-source software a prerequisite for the smartphone boom? iOS is not open-source, which arguably kicked-off the smartphone boom. Yes, Android is open-source (sort of). Facebook? No. Twitter? No. YouTube? No. Fortnite? No. And once again, RISC-V is an ISA, not even microprocessor IP, let alone "hardware".

Having pulled the article apart, the basic message is actually true. RISC-V is potentially disruptive. It has already taken over academia and the security research world. Obviously, in classic Innovator's Dilemma style, if disruption happens then it will start "at the low end", everywhere other than PCs, servers, and smartphones first. But don't forget all the skepticism around the Linux operating system. Today the cloud runs on Linux. Supercomputers run on Linux.

Semiconductors in India

The article on semiconductors in India has two big issues. One is that it completely ignores the large amount of semiconductor development that already takes place in India. Most (all?) major semiconductor companies already have significant development groups in India. I don't think that a list of the companies presenting at CDNLive India is necessarily the best way to survey an entire country, but this year there were presentations by: Intel, Google, Samsung, GLOBALFOUNDRIES, NXP, STMicroelectronics, Analog Devices, Qualcomm, AMD, Texas Instruments, Broadcom, Cyient, Open-Silicon,  Sanmina SCI Technology, Arm, Agnisys, L&T Technology Services, Mitsubishi, Agiliad Technologies, and Rambus. Those are all multinationals. But there were also presentations by Tejas Networks, Sankalp Semiconductor, HCL, Invecas, CDAC, ISRO, and eInfochips which are Indian companies. Oh, and Cadence presented too, of course. We have four engineering groups in India, in Bengaluru, Noida (Delhi), Pune, and Ahmedabad.

The bigger problem is that the article can't quite decide if "making" a semiconductor means designing it or manufacturing it. Here's the second paragraph:

Yet although many Indians work with computers, very few are employed in building them. All the components used to create Jio’s network were imported...Last year India imported $55bn of electronic goods. It exported just $8bn.

I think you could change "Indians" to Americans, "Jio" to Verizon, and update the numbers for the US market, and that would be true too, although nobody would open the sentence with "yet". Almost every component in your smartphone was imported, in particular, the big chips. They were manufactured in East Asia, but largely designed elsewhere—even in India!

The fact that India’s most celebrated industry depends entirely on imports in an era in which many countries are increasingly capricious about what goods they will allow to be exported makes some officials nervous. So India is attempting to build its own chips.

Not really. I think they would like to, but the investment to build a competitive semiconductor manufacturing industry is measured in tens or hundreds of billions. Just ask China that is investing over $100bn.

And the RISC-V confusion returns, with Indian flavor (masala):

Unlike the chip designs of Intel or Arm, which are proprietary, RISC-V designs are available to anyone with an internet connection to download free of charge, and to incorporate into their chip designs without a licence...CDAC has finished the design of its first RISC-V chip, and will soon start manufacturing it.

Once again, RISC-V is an ISA, not a processor design, although there are open-source processor designs out there. But CDAC (the Centre for the Development of Advanced Computing) is not going to be manufacturing it in India unless it is at SCL (Semi-Conductor Laboratory). SCL, part of the Department of Space, has a fab and builds some chips for space and military applications, but it has thirty-year-old technology. Wikipedia (not always the most reliable source^{citation needed}) says there are plans to upgrade their 6" fab to 0.25um (aka 250nm).

What India is actually attempting to do is to design some of their own chips.

Developing an indigenous semiconductor industry will be hard, however. India does have talented engineers, but only from a handful of elite engineering institutes.

India already has a lot of indigenous design expertise, just see my earlier list of CDNLive presenters. Engineering in India is not at all what, earlier in the article, The Economist calls "dull business-process-outsourcing and back-office management". But it is true that much of semiconductor design is being done by subsidiaries of American and European semiconductor and system companies, although there are Indian fabless semiconductor companies too. But despite the last paragraph of the article mentioning "Indian-made semiconductors", that is not happening anytime soon.

Again, like in the RISC-V article, the thrust of the article is true: some Indian companies are designing their own chips, but those chips will not be fabricated in India and so will remain "imported", probably forever.

CDNLive India

All those presentations at CDNLive India I mentioned earlier? They are now available online.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.