Flex Connections Lower Power Off the Top

People have always been looking for cost-effective ways to send chip-to-chip signals on a board at higher speeds while keeping power consumption down. While some investigate optical communication between chips and the like, Silicon Pipe (Mountain View, Calif.) has demonstrated the ability to send 10 Gb/sec signals from one end of a board to another using technology whose cost and power consumption is more consistent with that required of portable devices.

Last year, the company presented its Off-The-Top (OTT) technology, in which high-speed signals are routed to the top of the package and sent to other chips using an inexpensive, dedicated transmission line.¹ Then, the ability to send signals short distances without SERDES circuitry was emphasized. Since then, the company has demonstrated the ability to send high-speed signals from a SERDES chip through a 75 cm loop at 100 mV, rather than the standard 800 mV.

The loop included two 25 cm sections of flex transmission line that went from the SERDES chip to modified, off-the-shelf connectors. Another 25 cm section of flex functioned as the backplane between the connectors (Figure ). This was equivalent to two chips 30 inches apart communicating with each other as though they were next to one other. The power used by this chip was less than 2% of the power normally used in this type of situation.

Using dedicated transmission lines on flex, high-speed signals can be sent long distances at low power. (Source: Silicon Pipe)

From the viewpoint of the high-speed signals, this is simpler. Instead of being routed through a variety of material sets and impedences to a board that slows them down, they go straight to impedence-matched transmission lines. They arrive quickly and with much less degradation this way. According to the company, 3-D field solver models indicate that 20 Gb/sec is possible.

From an assembly perspective, this approach adds complexity, but not much. The company has explored the use of anisotropically conductive adhesives (ACAs) and simple tools to attach the transmission lines to OTT connections in its test vehicles, but soldering and separable connections remain options. Though the result has additional flex connections between chips and connectors, it is no more complex than some of the schemes proposed for sending optical signals between chips.

The power savings available, if the 100 mV voltage is any indication, could add more than enough value to a system to justify the additional effort required for assembly and rework. Also, there seems to be a wide margin in which to perform a trade-off between power consumption and signal integrity.

The creation installation of transmission lines on flex is an inexpensive proposition, but the rest of the scheme requires that chips and their packages be designed so that high-speed signals are routed to a different plane than the power, ground and low-speed signals. As it turns out, chip-package-board codesign is an emerging trend, due mainly to the needs of 90 nm IC technology.² The development of design tools necessary to use the OTT scheme may require little, if any, additional effort on the part of design tool vendors already in the codesign market. Conditions seem to be set to let people find out if the OTT scheme can follow through on its promise.

If it can, it may also deliver some time-to-market advantages. Though codesign of the chip, package and transmission line would be required to achieve the proper matching, the reduction and simplification of the I/O drive circuitry that is possible could actually make the IC designer's job simpler. The use of pre-emphasis or sophisticated algorithms to interpret degraded pulses could be reduced or eliminated. These simplifications would also leave more space on the chip for the IC designer to work with.

A scheme like this may be useful for another recent trend — distributed computing. It is not just a way to harness spare computing power for large tasks. The presence of special circuit blocks in leading-edge processors, along with the announced plans to make dual-core processors, can be viewed as instances of the distributed computing concept within a chip. A low-cost, low-power way to send high-speed signals between chips could allow a designer to distribute functions among a few chips, rather than use a single-chip implementation that might be unwieldy.

There are many multichip packaging options available now, and there are many technologies on the way designed to integrate disparate functions on one board or package. The widespread use of these technologies is an upcoming trend. Historically, though, it has usually been a simple, low-cost technology used in a low-cost application that has gotten large trends started. Though it remains to be seen what multichip technologies take hold, and in which market segments, the presence of a low-cost option like OTT is an indication that the market for multichip technologies is about to grow.