If you want to see the embedded industry out in force, forget San Jose and Boston – book a ticket to Nuremberg for the first week in March. This year numbers were slightly down on visitors (just under 16,000) but up on exhibitors – over 700 of them. And I think that this will be seen as the year in which the embedded world, more broadly, will be seen to have changed. One reason for this is 15 billion. Even in a period where governments and banks are lightly throwing trillions of dollars around, 15 billion is a big number. And it is a number that a lot of people were using at embedded world, being handed it by market analysts IDC earlier this year. Even Intel was using the number. So why? Well IDC has predicted that by 2015 there will be 15 billion embedded devices connected to the Internet. (Today, the company estimates, there are a mere 2 billion.) 15 billion is close to 2.5 times the estimated population of the planet.
Every one of those devices will have at least one programmable element inside, many of them will have more, most of them will be low cost and with short lives, so there will also be a significant replacement market on top of the growth market. The silicon vendors are licking their lips. And silicon giant Intel certainly wants a slice of the pie and has launched new members of the Atom family, including industrial temperature versions and a range of packaging options, in another attempt to get into the embedded market. (I know they have had an embedded group for 30 years – it says so in the press release – but they are not a name one instantly thinks of.)
The company is committed strongly to the embedded market and will be providing up to seven years life for some of the family members. (So not shooting at the defense market then.) Intel has clearly been working hard at building a community, as there was a raft of announcements of Intel Atom-based products from the likes of Kontron, Congatec and MSC. Green Hills announced Atom-targeted software, including the Integrity RTOS with virtualization technology and development and debugging tools, as did Wind River, as part of a larger announcement.
At the same time as announcing the embedded Atoms, Intel also announced that it is going to use an external plant to build Atoms and apparently will be licensing Atom’s IP. Now this is something new – clearly they are taking the war to ARM. And since they have revenue streams from a range of products, and since they already have an Atom ecosystem in place, unlike other entrants to the IP market, they can afford to wait while their IP works through the chain to generate real revenues. (Jim Tulley of Gartner explains the chain as:
You sell a license to someone, who designs a chip using the IP. This takes time.
They have to sell the chip to someone else, who designs a product using the chip. This also takes time.
They start selling the product in the market. Yet more time. Once the product starts to sell, the builder begins to buy real quantities of silicon, and then the silicon manufacturer starts paying the royalties to you. For volume revenues repeat this several times over.)
Just starting down this path is the ARM Cortex M0, a very small and low-power version of the Cortex M design. The M3 is now some 3 years old, and silicon is available from a number of vendors (Luminary produced yet another clutch of processors at the show, this time aimed at motor control and related areas. Motor control is another trigger for silicon salivation. This is discussed at the end rather than creating another level of nesting).
Since, with a 32-bit processor as small and cheap as the M0, silicon device manufacturers will be able to use it as the heart of very cheap devices (device pricing below 70 cents was mentioned several time for chips, including processor, memory and specialist peripherals), what is the future for 8-bit devices? Depending on whom you talk to, the answers varied enormously. Silicon Labs, for example, pointed out that the 8051-based 8-bit products have a range of tools and other elements in the ecosystem. Efficient 8-bit code can require a very small memory footprint, and 8-bit processors can draw very low power and are very cheap. So for a wide range of embedded applications, there is no contest. Luminary’s Jean Anne Booth thinks otherwise. Although the company is not going along the Cortex M0 route (they see the design as more suitable for integrated device manufacturers rather than fabless companies), she feels that the thumb instruction set and the tools available can produce very compact code, more efficiently than by hand coding.
ARM points out that 8-bit code is fine as long as you are happy to use only 128k bytes of RAM. If you need more than that, it has to be treated as external memory, with consequent overheads and inefficiencies. The perception that 32-bit development is complex is also a myth, since the developer can use many of the same tools as the 8051 developer. The killer reason for ARM is that code generation tools, which are going to be more and more important to build those 15 billion nodes, generate 32-bit code.
Microchip has a totally different argument, saying that word length is irrelevant. Instead, developers should choose the product that has the peripherals, speed and price to carry out the task they need. Their products all use the same development environment, run the same software with the same libraries, and even have the same pin out.
Conclusion: If this were a stock-tipping column, instead of something far more serious, I would make 8-bit a Hold, 32-bit a Buy, and 16-bit a Sell.
There was a lot more going on at the exhibition, and some of what I learned will be appearing in future articles. But one trend was very marked: tool companies are creating even more alliances to provide improved tool chains. One of these, which I will certainly return to, is Interested, being driven by Esterel, the France-based developer of modelling tools for high reliability systems. Interested is a European-Union-funded project to create a complete reference tool chain, using only European tools, for developing mission-critical and safety-critical applications. Although initially covering only European tools, the framework will be open, allowing other tool makers to plug in.
Why do I feel that this year’s embedded world was important? Because it marked a really strong play by Intel to topple ARM from its strong position. If they get it right, the face of embedded will be changed dramatically.
To avoid three-level nesting in the main article, I said I would return to motor control. This has been a regular topic at embedded world and other conferences for many years. The argument goes like this: add up all the electric motors in your home: refrigerator, washing machine, microwave, and every other device with a fan or pump. Throw in heating and air conditioning. Add in personal devices, hair dryers, things called, I believe, electric razors and more intimate devices. How about your office? Computers, printers, copiers? Public transit? The number of motors in your life, even outside manufacturing, is enormous. Think of the effect on the electricity distribution grid of reducing each motor’s energy use by as little as 10%. And efficient motor control circuitry can do much better than that. So again, licking their lips, silicon vendors have been developing specialist devices for at least ten years. There are two problems. The first is that the efficient control of motors is slightly more complex than just implementing a few algorithms and connecting up some interfacing circuitry. The second is that many of the products that use motors operate in very tightly cost-constrained markets. Even adding a 70 cent device could be an issue. So expect to hear more and more announcements of devices and development and prototyping kits.