Market Focus: Analog/mixed-signal

Though solid numbers are almost impossible to come by, the total value of mixed-signal devices is several billion dollars, according to Jim Feldhan, an analyst at Semico Research Corp., Phoenix.

The Semiconductor Industry Association groups mixed-signal ICs in two categories: custom, or special, analog circuits and digital ASICs, with worldwide revenue in 2000 reaching roughly $19 billion and $9.9 billion, respectively, according to the SIA.

How much of each of those fi gures is mixed-signal is uncertain, but the bulk is in the special analog segment, according to analysts.

This custom category has been under price pressure this year, but it hasn't been as bad as in other chip markets, according to SIA figures. The same is true for digital ASICs.

Market researchers and the SIA are working on ways to quantify mixed-signal-IC sales. The SIA is now counting a category called special consumer circuits under linear ICs. Special consumer circuits include mixed-signal chips for consumer, video, computer, and mass-storage applications-all growing markets for mixed-signal chips.

Besides analog-to-digital and digital-to-analog converters, some standard-cell devices have mixed-signal capabilities.

SIA figures put the worldwide standard-cell market at about $9.5 billion last year.

That category might include devices from Texas Instruments Inc. and STMicroelectronics Inc., which claim to have microperipheral circuits with mixed-signal capabilities, Semico's Feldhan said.

Microprocessors with embedded flash memory might also qualify as mixed-signal.

"TI will report a DSP core as a DSP core, even though it has memory and mixed-signal capabilities," Feldhan said.

DSPs are both a complement to mixed-signal and one of the crowd. "DSP doesn't live well without mixed-signal somewhere," Forward Concepts' Strauss said. "The chip itself is mixed-signal if it has onboard converters, and increasingly they do."

Microchip Technology Inc.'s PicMicro microcontroller is marketed as a digital device, but 30% of the circuitry is analog, according to Ron Cates, marketing manager at the Chandler, Ariz.-based chip company.

Memory, high-speed point-to-point serial links, and general interfaces are three of the most common functions for integration. "An area of high interest lately is the integration of the optical transceiver to its physical-layer neighbor," Power X's Weir said.

The line between mixed-signal and systems-on-a-chip methodologies is blurring.

Mor e and more so-called SoCs are going to have mixed-signal capability, particularly when production capacity is short, Semico's Feldhan said. "You may want to spend the extra money to get an integrated chip, even if it's not an absolute system requirement, because you'll have fewer chips to buy."

Mixed-signal elements can be included on an SoC, said John O'Boyle, director of marketing at Samsung Semiconductor's Cubic Solutions Group in San Jose, a maker of digital ASIC and mixed-signal SoCs. "We typically include a processor core with A/D converters, phase-locked loops, and RF blocks."

Cypress MicroSystems has developed a reconfigurable microcontroller it calls a programmable system-on-a-chip (PSoC). This includes flash memory, SRAM, and programmable arrays of analog and digital system functions. The device allows designers to dynamically configure and reconfigure the MCU core late in the design stage. This is a first, according to Cypress MicroSystems.

Multilink Technology Corp. calls its 10Gbit /s serializer-deserializer (SerDes) an SoC: a multiplexer with integrated clock multiplier unit and demultiplexer with integrated clock and data recovery.

Decision time

Implementing mixed-signal technology involves a number of trade-offs. For one thing, not everything needs to be integrated. The cost of integrating functions on one piece of silicon needs to be balanced against the volume of the end product and performance requirements.

"In a high-performance situation, where having a lot of signals going off chip and on chip would slow the performance, you would decide to integrate it. You don't care if it costs 10 times as much because you need the performance," Feldhan said.

But if the design process takes too long to integrate the chip relative to the product lifecycle, the OEM may go for less-expensive discrete analog and digital chips, he said.

"You're basically trading performance vs. cost," said Richard Chesson, director of business development, Americas, for the telecomm unications, peripherals, and automotive group at STMicroelectronics, San Jose. "And that cost can be in die area, but it can also be pushing you into a leading-edge technology-maybe one that's so leading edge it's not quite ready or one that's capacity-limited."

Trying to integrate everything into a part can defeat the purpose of a mixed-signal chip, said Gerry Jurkovic, director of PHY/PMD marketing at Multilink Technology, Somerset, N.J. "A super part may get into very high power dissipation, high cost, a larger die size, and so on. It all depends on the variability of requirements in the marketplace."

The OEM is essentially making a trade-off between flexibility and integration, Jurkovic said. "The advantage of integration is potentially lower cost and potentially lower power. But there's a trade-off also in terms of being able to select the best-performing products in the market."

That means that by not integrating, the OEM can mix and match analog and digital parts from different vendors to suit its needs. Users are getting more comfortable with multi-die packages, which can be more cost-effective than an integrated solution.

Cost to the OEM is not only in the mixed-signal chips themselves, but what they offer for the overall system, according to Power X's Weir. An OEM building, say, a line card, should ask what the advantages are of having a mixed-signal/analog function integrated in the general chipset solution, he said.

"A block of digital ICs would be smaller and somewhat cheaper than with an integrated analog function," Weir said. "But there are hundreds of analog serial links that need to be implemented somewhere. Now the OEM has to buy all these digital ICs and then buy all the external analog transceivers that go around it."

Levels of integration

Integration is usually defined by economics rather than technology, said Alun Roberts, director of marketing operations for high-performance analog products at TI in Austin, Texas. The size of the customer's market oft en determines the need for integration, he said. Handsets, for instance, are high volume and highly integrated with analog, digital, power management, and RF capabilities.

Wireless basestations, on the other hand, emphasize performance, and "the more performance you get out of a basestation, the fewer you have to build," Roberts said. These products will be less integrated and have more powerful chips or multichip solutions.

Integration always means some level of performance compromise, often a minor one, Roberts said.

Memory is always a trade-off because it requires more mask steps, creating more cost to the OEM, Samsung's O'Boyle said.

OEMs need to weigh the performance desired against the processing parameters.

Analog design rules have held at around 0.35- and 0.25-micron processes, while digital is at a 0.18-micron and even 0.13-micron level. Combining these on a single chip involves a complex process. As line geometries get smaller, the voltage is reduced, so it becomes tougher to i mplement mixed-signal technology. A 0.18-micron process can support 3.3V, but it can't support 5V.

The high-performance analog process generally gets more difficult as you reduce the supply voltage because you have less voltage range to work with, said John Hussey, vice president of the high-speed converters group at Analog Devices Inc., Norwood, Mass. So if it goes from 0.35-micron to 0.18-micron and even 0.13-micron, it gets harder to increase the analog performance at a lower supply voltage, he said.

With smaller line geometries, the cost goes up dramatically, Hussey said. "The analog part costs two-and-half times to do in 0.18-micron as it did in 0.35-micron because it doesn't shrink as much as digital," he said. "So if the analog is a significant part of the chip area, you're paying two-and-a-half times as much for that part of the chip," he said.

OEMs start designing a mixed-signal chip by defining the analog requirements, but digital circuitry is characterized first in the design, said Pau l Lombardelli, director of IC design at Power X. "The digital portion accepts the minimum geometry you can use," he said. "For digital, you just build a standard cell, qualify it, and replicate it. For analog, you have to handcraft every transistor."

Analog is typically 30% to 40% of the mixed-signal design. But increasingly the digital portion is incorporating some sort of processor as well as memory, STMicro's Chesson said. As a result, power consumption also becomes part of the price/performance equation, he said.

CMOS vs. GaAs or SiGe

CMOS remains the bread-and-butter process for mixed-signal, and has even gained some advantages over the more exotic and more expensive alternatives silicon germanium (SiGe) and gallium arsenide (GaAs), according to Lombardelli.

"You can build almost anything in a standard CMOS process," Lombardelli said, adding that the speed a CMOS process can handle has increased about 10 times in the last few years.

Implementations such as integrating a mass ive processor with lots of memory and complicated high-speed analog power previously had to be done in BiCMOS or gallium arsenide. Very-high-speed links and RF amplifiers that were always processed in GaAs and SiGe are now possible in straight CMOS, Lombardelli said.