Chips that reconfigure on the fly

Future Watch by Jaikumar Vijayan

MARCH 22, 2004 (COMPUTERWORLD) - Imagine a handheld device that combines the functionality of a cell phone, a PDA, an MP3 player, a digital camera, a television and a half-dozen other devices.

Whatever the merits of such a gizmo, for the moment at least, building it is still infeasible. The separate microprocessor logic needed for each function would require far too much real estate and energy to fit into anything resembling a decent handheld.

Proponents of a long-held concept called reconfigurable computing think they may have an answer to the problem.

Reconfigurable computing involves processor hardware capable of automatically adapting to changing application needs. Unlike the permanently etched circuitry on current-generation static microprocessors, the wiring on reconfigurable processors can be redrawn on the fly by software- or hardware-based microcontrollers, to match the function being performed. By opening and closing the millions of logic gates on such chips, their circuitry can be changed so as to perform signal processing one instant, for example, and an encryption function the next.

Because reconfigurable computing chips do away with a lot of the redundancy and overhead found on static processors, they also consume less energy while delivering greater speed.

For instance, a single such chip in a cellular handset could configure itself to search for a local base station, then establish its identity on the cell and finally send or receive calls - tasks that today require multiple chips.

``What you are trying to do is to change the hardware to match the problem at hand,`` says John Watson, co-founder of QuickSilver Technologies Inc., a developer of adaptive computing technologies in San Jose.

Such malleable chips represent a fundamental shift in microprocessor design, says Nick Tredennick, a former microprocessor architect and editor of the Gilder Technology Report in Great Barrington, Mass.

``All the microprocessors that we have today are basically 30- or 40-year-old designs,`` says Jim Turley, an independent analyst in Pacific Grove, Calif. ``We are fundamentally making the same computers our grandfathers did, even though silicon technology has improved dramatically.``

Though it might sound radically new, the concept of adaptive computing architectures has been floating around in one form or another for some time. Indeed, several companies - from relatively unheard-of start-ups such as QuickSilver and GateChange Technologies Inc. to the likes of Intel Corp., Motorola Inc. and Infineon Technologies AG - have efforts under way in the area of reconfigurable computing.

For example, QuickSilver offers an integrated circuit technology that dynamically changes at runtime to create the hardware needed for different applications. Someday, such technology could form the basis of a universal cell phone or be used in flexible automotive electronics components, Watson says. Similarly, Intel`s efforts in this area include a project to develop software-defined radio that uses software to automatically reconfigure all the hardware logic on a chip, with no mechanical switching of circuitry.

The idea is to eventually build a general-purpose wireless communications processor that`s capable of executing multiple standards and protocols and offers better performance than a dedicated application-specific integrated circuit (ASIC), says Steve Pawlowski, an Intel research fellow focused on reconfigurable logic.

Such efforts have gone commercially unnoticed for a long time, but they`re going to be crucial in tomorrow`s untethered world, Tredennick says. Conventional processors were designed for systems that were largely static and derived their power from wall sockets, he says. As a result, the focus of most early microprocessor design was on price/performance per unit - squeezing out ever-faster clock speeds from a piece of silicon while keeping unit costs reasonably low.

With the advent of laptops and notebooks, the focus shifted to microprocessor power levels as well, Tredennick says. As more devices start going mobile, there is going to be an even greater focus on price/performance per watt of power consumed, he says.

Fixed-circuit microprocessors are quickly reaching certain physical limits that will make it extremely hard to maintain the power/performance balance for much longer, making reconfigurable architectures inevitable in the untethered world, Tredennick says.

But getting there isn`t going to be easy, everyone agrees. In fact, few expect to see widespread use of dynamically reconfigurable chips within five years.

Pawlowski predicts that it will take at least until 2007 for Intel`s software radio efforts to become commercially viable. ``We have to first prove internally that there is a significant product benefit in terms of costs, power and area. Our mobile power people are expecting longer battery life and the flexibility to handle multiple protocols,`` he says. ``We are trying to find the perfect fit.``

Also, technologies such as ASICs and field-programmable gate arrays (FPGA) already offer some of the same benefits being touted by adaptive architectures. ASICs are special-purpose chips and are more efficient at doing specific tasks than general-purpose microprocessors. FPGAs are devices whose hardware logic can be reprogrammed in the field to accept hardware updates or bug fixes. But neither ASICs nor FPGAs offer the flexibility that adaptive architecture promises. Both technologies are also relatively expensive and, in the case of FPGAs, too power-hungry.

Building adaptive chips can also be difficult. Writing the programs that instruct each of the millions of logic gates in such chips to open and close so as to alter the circuitry on the fly can be incredibly complex.

The hardware design tools to do these sort of tasks are only just emerging, Pawlowski says.

Also, Turley says, ``most people haven`t really wrapped their heads around how to create and manage computers that change on the fly.`` Nor is there a full understanding of what applications might truly benefit from the approach, he says. For the moment, mobile devices present the biggest opportunity for adaptive computing.

``Someone is going to have to build a very high-volume application for the case to be really compelling,`` Tredennick says. ``Nobody has done anything that is really [commercially] successful`` so far. But that could begin changing soon, he says.