PORTLAND, Ore. — This follow-up to the EE Times story TI Obsoletes FPGA — which generated a torrent of comments both pro and con — is our attempt to clarify your outstanding questions.
Texas Instruments (TI) has thrown down the gauntlet to the field programmable gate array (FPGA) community — mostly Altera and Xilinx — by starting a family of special-purpose processors that combine multicore ARM processors with multicore digital signal processors (DSPs) and multiple programmable hardware accelerators. Can they pull it off? An FPGA is like a blank-slate that can solve any high-speed computational problem, but at a high-price, difficult programming and a waste of resources (not every gate is used in the vast majority of FPGA designs).
On the other hand, TI will have to create an expanding family of special-purpose chips to address all the niche markets that FPGAs address. Can it be done? We asked TI, Altera and a bevy of analysts what their opinions were and summarized them below. (FPGA-maker Xilinx was contacted too but said "no comment.")
Die shot of an Altera Stratix IV field-programmable gate-array (FPGA) with integrated 11.3 Gbps transceivers.
"In my opinion, the TI product offers an interesting prospective, because it can target existing solutions on two fronts: on the FPGA side, the solution offers the lower cost and lower power consumption that underpins the SWaP-C (size, weight and power/cost) requirements for both military and industrial applications while offering comparable performance. In this respect, the solution can not only provide an avenue for more portable solutions, but also offered as an effective replacement for FPGAs in existing equipment," said Asif Anwar, director of Strategic Technologies Practice (known in the U.S. as Strategy Analytics Inc., Boston) "The second front could be the GGPU [general graphics processing unit] segment."
Altera HardCopy II 230 ASIC (application specific integrated circuit) which is compatible with Altera's Stratix II FPGAs.
According to Anwar, COTS (commercial-off-the-shelf) offerings based around GGPUs have also been offered as alternatives to FPGAs and ASICs (application specific integrated circuits) especially for applications where there is a large amount of data that needs to processed in the same way, for example SAR (synthetic aperture radar) imagery. The TI SoC, said Anwar, can also target this area while providing more powerful processing capabilities making it an ideal solution for customers that would prefer the performance attributes associated with FPGAs but are having to use GGPUs as a 'good enough' solution because of cost constraints.
"On the GGPU, it is also worth highlighting the fact the there is an inherent limitation where more flexible processing capabilities are needed making them less suitable for applications such as electronic warfare [EW]. The TI solution on the other hand should be well suited to handle the additional complexity in processing requirements associated with EW systems. Will the TI solution make FPGAs obsolete? Off course, there will still be markets where FPGAs will remain the potential 'go-to' solution in terms of the performance/cost argument, for example, in the space sector where FPGAs compete against ASICs. I don’t think it is something that will happen overnight, but if TI can offer a family of solutions, then it will offer a credible alternative that looks capable of eroding the market position held by FPGAs."
On the other hand, some analysts were practically spewing smoke from their ears over TI's attempt create a family of heterogeneous processors to obsolete the FPGA, even if only in certain applications.
"The short answer is that TI is far overstating their case," said Linley Gwennap of The Linley Group (Mountain View, Calif.)
The Xilinx XC2064 FPGA was first to use a LUT (look-up table.)
Linley himself referred us to one of his analysts for more details, Jag Bolaria, who explained Linley Group's conclusions in a little more detail.