A Chip off the New Block

Matt Chang is a hardware design engineer at Opto 22, the Temecula, CA-based automation hardware and software manufacturer. As such he has a hand in the development of new products, and something about embedded control chips used to bother him. The chips would have features that he, or any automation hardware designer, would never use.

By Hank Hogan for Control Engineering May 1, 2007

Matt Chang is a hardware design engineer at Opto 22, the Temecula, CA-based automation hardware and software manufacturer. As such he has a hand in the development of new products, and something about embedded control chips used to bother him. The chips would have features that he, or any automation hardware designer, would never use. Those extras wasted die space while increasing cost and power consumption.

He doesn’t worry about that anymore. “In a lot of the newer products, the semiconductor industry is almost beyond that now,” says Chang.

Advances in semiconductor manufacturing technology, he explains, have resulted in transistor densities so great that extra features take up very little room. On top of that, what was an extra before—such as an Ethernet or wireless interface—is now essential. For Opto 22, which typically designs using off-the-shelf semiconductors, chip costs have dropped significantly and the level of integration has increased considerably.

That situation and those benefits aren’t unique to Opto 22. A look at several semiconductor manufacturers shows where embedded chips are today and where they could be tomorrow.

Inside Intel

Everyone who hasn’t been hiding in a cave knows that Intel Corp. of Santa Clara, CA, manufactures microprocessors. What isn’t as well known is that the world’s largest chipmaker was in the embedded systems market before it was in the PC market, according to Phil Ames, Intel’s segment marketing manager for the embedded and communications group.

In January, Intel announced a breakthrough in the fabrication of 45 nanometer transistors, devices so small that 30 million fit on the head of a pin. The company expects to begin production using the process in the second half of this year. Little more than a half decade ago, such an announcement wouldn’t have meant much to embedded chips because the lag between embedded and the most advanced processors was running as much as 24 months. That’s no longer the case, notes Ames. “We’re seeing almost zero lag now.”

He points to two cases in 2006 where an announcement of a multiple core product for desktop use was followed within weeks with a launch of an equivalent embedded chip. Such advanced technology benefits automation applications in a number of ways.

For example, a chip with multiple independent processors can dedicate one core to running a robot under a real-time operating system while the second core runs enterprise applications. Thus, essential control isn’t starved for processing cycles or memory. An added advantage, notes Ames, is that spreading out the workload over a greater die area brings the power consumption down. Real-world demonstrations have shown that, when compared to a single processor, two cores can run twice the applications at 40% less power—an important advantage for embedded applications.

Not every Intel chip will appear in an embedded version. The company validates both a chip and its fabrication so that Intel can guarantee five to seven years of manufacturability and support. Intel does this because industrial applications have a longer life than consumer products, and it wants to make sure that selected products will meet the longevity requirements before committing a multibillion dollar fabrication facility to the product’s support. “Once the validation is complete and we know we can guarantee these products for a long life, then they go on a long life or embedded roadmap, which is slightly different from our desktop offerings,” says Ames.

Meanwhile, across town

Advanced Micro Devices Inc., or AMD, is Intel’s Sunnyvale, CA-based rival. It too has embedded chips, notes Jeff Chu, division marketing manager for the embedded computing solution division at AMD. The lag between enterprise processors and embedded versions running on the equivalent process is now about six months, down from a much longer delay in the past, he says.

Chu says that, for whatever products it introduces, AMD leverages the latest technology so as to address the needs of the embedded market. These include such issues as longevity of the product and its power consumption. For example, he notes that AMD has released a dual-core Athlon chip aimed at higher-end embedded applications with a nominal power rating of 35 Watts. However, the processor also has power states as low as 12 Watts. “You can set the speed so that you can keep the power consumption down,” says Chu.

While such processors might be found in single-board industrial computers, AMD also offers its Geode line, which is very compact and very low power. The processors can operate at 1.3 Watts typical power while running at 433 MHz. Their capabilities include the ability to natively run all Microsoft Windows and Linux-based applications and access to all 32-bit X86 software, among the most diverse and widely available. The latest version of the Geode line runs at 600 MHz and 2.6 Watts typical power.

Not every single new chip from AMD will appear on the company’s embedded product roadmap. Part of this is because the company is focused on providing embedded products for five years, which entails a relatively long commitment to a given process and product. Beyond what customers desire in terms of longevity, there can also be issues with certification bodies, such as the Federal Aviation Administration. In some cases, once a particular system is approved, it and the parts within it cannot be changed. Thus, AMD doesn’t want to enter into an agreement to produce a part without considering such factors.

Your design

Besides well known names like Intel and AMD, there are several other embedded chip players. One such is Austin, TX-based Freescale Inc. Spun off from Motorola in 2004 and now privately held, Freescale has a decades-long history in embedded controllers. While not investing as much as some larger firms do, Freescale has been spending over a billion dollars a year on various semiconductor processing research and development projects.

Jeff Bock, mobile product marketing manager at Freescale, notes that the company has been driving down the power consumption and price of its products. For power consumption, the company’s processors have new chip level architectures that allow customers to select which peripheral circuitry within the chip to turn on or off. Originally developed for mobile applications, this capability is being employed more and more often in embedded applications.

In the embedded space, Bock sees a trend away from 8-bit and 16-bit chips and to 32-bit devices. As for why this is happening, he points to Controller Area Network (CAN) implementations, as well as to the emergence of such wireless protocols as ZigBee, which is a low-power implementation intended for short range use. The desire to take advantage of such communications protocols often prods people to abandon older processors and move up. “Communication is often what’s driving people to move to 32-bit and higher performance levels,” says Bock.

Finally, there are those that devise their own solution and don’t depend upon a general purpose embedded processor. Milwaukee, WI-based Rockwell Automation, for example, is an automation supplier known for its Logix programmable automation controllers. The company isn’t a semiconductor supplier for the mass market, but it does design its own chips.

Scott Tenorio, manager of controller strategy for Rockwell Automation, notes that there are several reasons for this roll-your-own approach. Thanks to silicon foundries, it isn’t that expensive to do a custom chip. Beyond that, this method allows the company to ensure the chips work well on the bit level and it enables Rockwell to make the chips as compact and as power efficient as possible.

This isn’t to say that Rockwell will only use its own designs. It all depends upon the particular application and what is deemed to be the best solution. “We are definitely viable to going off the shelf,” says Tenorio.

Into the future

Whether based on a custom design or a general purpose one, all embedded chips exist within and benefit from general semiconductor manufacturing technology. That universe has long lived in compliance with Moore’s Law, the observation first made more than 40 years ago by Intel founder Gordon Moore that the number of transistors on a chip doubles every few years.

Historically, this has meant a 0.7x scaling of the linear dimensions of critical chip features every three years, says Bob Doering, a senior fellow at Texas Instruments (TI), Dallas. That scaling has contributed to companies like TI being able to offer embedded controllers at ever lower prices and with ever more capabilities.

Starting in the late 1990s that trend sped up, with the scaling time dropping to 2.5 years. The pace seems to have slowed down within the last few years, and is closer to the historic norm. What’s unknown, though, is what will happen in the future. Doering thinks that the pace of innovation won’t slow down until the end of the decade, or sometime after 32 nm processing is introduced.

Beyond 2010, though, things get cloudy. It may be, to paraphrase a saying, that the technology is willing but the economics are weak. It may simply cost too much and be too big a risk, especially if fundamental changes in photolithography or other fabrication techniques are needed.

Doering, though, cautions that it is not wise to discount the ingenuity of those in the industry. The current processing savior, immersion photolithography, is a semiconductor implementation of an idea that has been around in microscopy for centuries. So solutions to looming problems may be found and Moore’s Law may march on. As Doering says, “Our engineers are just amazingly clever in coming up with new wrinkles.”