X-EMI readies chips to cut EMI by 20 dB

A startup with roots in the network-processor company SiTera Inc. has demonstrated a unique solution to electromagnetic interference problems at a vertical EMI show in Chicago.

X-EMI Inc. (Austin, Texas) is using a proprietary method of spectral diffusion to modify system clocks in order to reduce EMI from a variety of sources.

Steve Sheafor, a SiTera founder who serves as chief technology advisor to X-EMI, said that the original single-channel transceiver shown to OEM customers this month would be followed quickly by quad and octal versions.

Unlike spread-spectrum clock concepts emerging from the PC community in recent years, the X-EMI solution does not change clock frequencies, and can be designed in to legacy systems.

Nevertheless, X-EMI is focusing primarily on designs for next-generation buses like PCI Express, since EMI problems expand as the square of the system frequency increase. There will be applications in existing industrial buses, but "as your bus frequency exceeds 100 MHz, what is advantageous becomes a necessity," Sheafor said.

Ideas were generated by several Rice University researchers, though some of the original concepts came from the University of Houston. Work progressed from initial concepts in early 2003 to early 2005, when former SiTera executive Larry Woodson was brought in as chief executive officer to bring products to market.

Sheafor said that the EMI technology is as universally applicable as bus and channel technologies like those supported by Rambus, though X-EMI's strategy differs in two important ways: the company will focus on sales of chips rather than intellectual property, and the level of complexity of spectral-diffusion is much lower than the bus technologies developed by Rambus, so that a smaller design team can keep products updated.

The first such product is the XM1001, a simple standalone oscillator and frequency synthesizer with differential and single-ended I/O. The device provides an interface between the X-EMI XClk signal and a normal clock signal. At a transmit end, the device converts a system clock signal into a very low EMI signal. When used as a receiver, the XClk signal is recovered and converted to the normal system clock. Current 28-pin devices will move to 8-pin designs in the future.

"The key to the patented technology is auto-sensing to tell whether we are encoding or not," Sheafor said. "These devices can be used within existing systems, with almost negligible footprint."

In X-EMI's Optimized Spectral Diffusion, the source signal is multiplied by the output of a digital noise source. The resulting clock is then sent to a destination and multiplied by an identical noise source. X-EMI uses its own analytical algorithms to create efficient spread spectrums, with very low peak EMIs.

Unlike native spread-spectrum clocks, the X-EMI method does not dither the signal, and affects all higher-order harmonics of a signal the same.

Chief technology officer Ken Egan has backgrounds in cryptography and graphics, and his work on algorithms demonstrated 6 dB reductions in EMI at University of Houston, now improved to more than 30 dB in some demonstrations. The algorithms can be implemented in a solution that is almost fully digital, with only small analog blocks.

"The beauty is that this can be implemented in simple circuitry incorporating thousands of gates," Sheafor said. "We've implemented the first XM1001 in quarter-micron CMOS."

Upcoming four- and eight-channel versions can be implemented just as generically as the single-channel device, though X-EMI will emphasize pin compatibility for specific environments. The quad XM1004 will follow Intel DB400 standards, and will be compatible with ICS DB104 and Cypress CY28400 clocks. The XM1008 octal buffer will follow Intel DB800 standards.

- Loring Wirbel

EE Times