Frequently Asked Questions
Download Arrix FPOA Overview
4 Page FPOA overview that includes: architecture information, algorithm support, development system, design flow and target applications.
2. What is the difference between an FPGA and an FPOA?
FPGAs (Field Programmable Gate Arrays) and FPOAs are both field programmable devices, but there are significant differences between them. In certain high-end markets such as high performance video and image processing, FPGAs and FPOAs are competitors; in other markets, FPGAs and FPOAs are often complementary solutions. Many times, a designer will have a choice between many large FPGAs or a small FPGA and an FPOA. Although every case is different, MathStar has typically been able to provide a 50% cost reduction over an all FPGA solution.
Areas where FPOAs are better:
a. The advertised operating frequency of an FPGA is around 500-600 MHz for the latest FPGAs. However, the actual operating frequency of a real design is often 250 MHz or less. This is because the process of porting a design to an FPGA and making all the interconnect timing work, called “timing closure”, tends to force the designer to reduce the actual operating frequency (“fMAX”). In contrast, the operating frequency of an FPOA is deterministic and can be as high as 1 GHz. Said another way, a 1GHz FPOA will always run at 1GHz
b. Many algorithms operate on “samples”. A sample is usually some number of bits greater than 8. For example, an A-to-D will sample an analog signal and represent it by many discrete values over time, each sample being some number of bits. Modern image and signal processing uses these samples as the input data stream. Because the FPOA’s fundamental operation is at 16 bits, it is much faster and more efficient at operating on samples. By contrast it is less efficient operating on individual bits.
c. FPGAs use memory bits for programming and tri-state buses for interconnection. Even when only a portion of the entire device is used, power is consumed for the entire chip. In contrast, on an FPOA, unused Silicon Objects are shut down. This makes a 130nm FPOA about the same power as a 65nm FPGA chip for the same performance and application.
d. The programming time for an FPGA takes seconds, particularly for the larger devices; it takes milliseconds to program an FPOA.
Areas where FPGAs are better:
a. An FPGA can be programmed at the gate level with arbitrary bit precision and functions, making it more flexible than an FPOA. Just as an FPOA is much faster and more efficient at operating on samples, an FPGA is more efficient (but not faster) at operating on individual bits.
b. FPGAs are programmed via a design flow that usually supports HDL synthesis. The FPOA does not support synthesis design flows at this time. The design flow for the FPOA requires the designer to map the algorithm to FPOA objects, which may not be a familiar process. However, once the algorithm is mapped, the FPOA performance and timing is deterministic. This is in contrast to an FPGA where there can be repeated design iterations to achieve an acceptable operating frequency. Download Arrix FPOA Architecture Guide
