The question of reconfigurability
The concept of reconfigurable computing initially emerged in 1985 when Xilinx's first data books said reconfigurable logic on the cover. Although the product was envisioned as a glue logic replacement, a few initial customers saw the value of using the Xilinx product also as an ASIC prototyping tool. It was not until 1989 that Stan Baker of EETimes coined the term FPGA, and Xilinx refocused their product and reworded their data book accordingly.
|Fixed: Non-Configurable -e.g., Pentium, PowerPC, ARM9, etc. Design-time Configurable -e.g., ARC, Tensilica, MIPS, etc. Procedure Reconfigurability -Context Switching in Software Structure Reconfigurability-Occasionally Reconfigurable -e.g., FPGAs, PLDs ††††††††††† -Dynamically Reconfigurable Logic -DRL: within a few cycles††††††††††|
Although memory-defined FPGAs are truly reconfigurable in terms of their circuitry and logic implementations, some argue that they are only “occasionally” reconfigurable, with reconfigurability times measured in several minutes to as little as a few seconds, depending on the amount of logic to be changed.
There are system-level products that would benefit from even faster reconfigurability, either on demand or as the result of detection of something that requires a logic change.
With logic changes implemented in the sub-millisecond range, Dynamically Reconfigurable Logic (DRL) allows the system hardware to dynamically adapt its architecture on-the-fly (even on each clock cycle for some implementations) to handle a new task.
The application that most people look to for such real-time reconfigurability is Software-Defined Radio (SDR). Basic SDR may be as simple as switching between two well-defined radio waveforms, each requiring unique logic.
The ultimate goal of SDR, however, is to examine the available radio signal types (waveforms) detected in a given vicinity, and automatically configure the radio’s hardware logic for desired reception (and perhaps transmission) totally in software, with no mechanical switching of circuitry. Some define this as cognitive radio, a superset of SDR.
Although one could argue that simple context switching from one set of algorithms to another can accomplish the job, that usually involves a duplication of memory and multiple circuits that are physically switched in or out, as is the case with current dual-band and tri-band cellphones. One cellphone maker’s view is that DRL promises to be more flexible than ASICs, and more energy-efficient (longer battery life).
The vision for an SDR cellular base station is for the DSP engine to dynamically switch from one cellular air interface standard to another, as may be the case for serving a number of both CDMA 1x-EV and GSM/GPRS/EDGE users in the vicinity. In practice, it is unlikely that mainstream CDMA operators like Verizon or Sprint would ever want to share operation with GSM-centric operators like Cingular or T-Mobile, but Mid-Tex Cellular, an independent operator that garners most of its revenue from subscribers roaming in West Texas, has implemented Vanu Incorporated's basic SDR base station for both GSM and CDMA operation.
On the horizon is WiMAX broadband wireless access (BWA) technology, that many envision as a technology that must co-exist with Wi-Fi and/or the several flavors of cellular. PicoChip Designs, for example, is working with a Korean group to implement their picoArray DRL chip in WiMAX/WCDMA base stations.
But, SDR is not the only application for DRL. Reliability through redundancy is another application, with dynamic bypass of a failed or faltering processing element. There are companies pursing this approach for high-reliability applications, likely for automotive, aircraft and military markets.
Others see DRL as the perfect solution to dynamically switch base logic required of the many different video codecs (MPEG-2, MPEG-4, H.264, VC-1, etc.) required of future personal video recorders, and even at head ends of cable networks and upcoming IPTV (Internet Protocol TV) systems that are beginning to roll out from U.S. telephone operators.
But, DRL is not easy. The streets are littered with companies that tried and failed. Some had good ideas, but ran out of money. Others ran into insurmountable technical problems. But, hope springs eternal. We at Forward Concepts count over 20 chip companies that are presently designing or actually offering DRL chip products. As with all chips, the market will decide.