Morpho Technologies and picoChip both recently introduced reconfigurable-processor-based solutions targeting WiMAX applications. The Morpho solution, which targets terminal applications, is a combination of software and licensable hardware IP for use in custom chips. The Morpho hardware IP is based on the “MS2,” Morpho’s second-generation reconfigurable processor core. Morpho’s hardware IP also includes a coprocessor for Viterbi and turbo encoding and decoding, as well as an interface for radio frequency hardware.
The MS2 architecture consists of an 8 2 array of data paths. Although it contains 16 data paths, the MS2 has a single thread of control, and it issues only one instruction per cycle. This single instruction can be dispatched to any number of data paths. After the instruction is dispatched, the data paths continue performing the specified operation until they are loaded with a new instruction. (From a high-level perspective, the operation of the MS2 is similar to that of a vector processor. To learn more about vector processors, see BDTI’s analysis of the Telairity TVP400 in the September 2003 DSP Insider.)
The MS2 is expected to achieve a worst-case clock rate of 250 MHz in the TSMC 90G process, which is a fairly low clock rate. However, Morpho says that the MS2 will use each clock cycle efficiently. For example, Morpho claims that the MS2 can complete a 1024-point FFT in 3080 cycles. In comparison, TI claims that its ’C64x+ can complete a 1024-point FFT in 3878 cycles. Mopho claims to have an even larger advantage in energy efficiency. When processing an FFT, for example, Morpho claims that the MS2 consumes about one-third as much power as competitors like the StarCore SC1400 and the TI ’C55x. (BDTI has not verified any of these performance claims.)
Perhaps Morpho’s most important feature, however, is flexibility. Transceivers in wireless terminals generally rely on a combination of a DSP and hardwired logic. The hardwired logic is typically needed to support demanding front-end processing. This hardwired logic is often inflexible, making it impossible to use the same hardware for different wireless communications standards. In contrast, Morpho claims that its fully-programmable approach gives it the flexibility to handle a wide range of wireless applications.
Of course, programmability is only an asset if it is practical to program the MS2. Although the MS2 is less complicated than some architectures used in wireless communications, it is still a relatively complicated design. Hence, system developers who want to program the MS2 will need good tools that support their efforts. BDTI has not evaluated Morpho’s tools, so it is not clear whether Morpho has met this goal.
Another challenge for Morpho will be convincing potential licensees that it has staying power. Morpho has not announced any licensees, and until it does so, potential licensees are likely to worry about Morpho’s long-term viability.
Like Morpho, picoChip recently introduced reconfigurable solutions targeting WiMAX applications. There are some interesting similarities between the offerings, but there are two key differences: First, picoChip is selling chips instead of licensing hardware IP. Second, picoChip’s offerings target line-powered applications, not portable applications.
The new picoChip offerings include three chips: the PC202, PC203, and the PC205. These new chips use the same basic architecture as PicoChip’s first-generation chip, the PC102. Like the PC102, each PC20x chip contains a large number of DSP cores. The main difference between the two generations is the number of DSP cores. While the PC102 contains 308 DSPs, the PC202 contains only 198 DSPs; the PC203 and PC205 each contain 248 DSPs. Not surprisingly, the PC20x parts cost less than the PC102. For example, the PC202 is priced at $44 in 10,000-unit quantities, compared to a price of $125 for the PC102.
In addition to reducing the number of DSP cores, picoChip also modified the accelerators used in the chips. The PC102 contains 14 small hardware accelerators; in contrast, each PC20x chip contains 5 larger, more powerful accelerators. The other major difference between the chips is that the PC202 and the PC205 each contain an ARM9E core. In contrast, neither the PC203 nor the older PC102 contain an ARM core.
As with Morpho, a key selling point for the picoChip solution is its programmability. However, the PC20x chips are quite complicated, and programming them is likely to be a challenge. BDTI does not have any first-hand experience with these parts, so it is difficult to know how significant these challenges are. (Like Morpho, picoChip offers WiMAX software for its processors—and picoChip also offers software for WCDMA applications.)
It is interesting that Morpho and picoChip have taken similar approaches: both companies target a narrow range of applications, and both companies offer off-the-shelf software for these applications. Perhaps the companies have learned from the failures of earlier companies that offered reconfigurable processors. Until recently, vendors of reconfigurable processors typically targeted a wide range of applications. Due to this lack of focus, these vendors rarely offered strong support for any one application—and the vendors may have failed to gain traction as a result. With their narrowly targeted solutions, Morpho and picoChip seem better poised for success.
The Morpho solution is expected to be available in July. Morpho has not disclosed licensing costs for the solution. The PC20x chips are available now. The PC202, PC203, and PC205 are priced at $44, $73, and $99, respectively, in 10,000-unit quantities.
Add new comment