1) FPGA have a faster time to market.
2) Compared to ASICs, FPGA are reprogrammable.
3) High-end FPGA can match the performance of an ASIC.
For these specific reasons, a few start-up companies specializing in deep learning applications are now turning to FPGA over an ASIC.
The first company is DeePhi and is a relatively new company. Founded in March 2016, DeePhi (short for the phrase “discovering the philosophy behind deep learning computing”) believes an FPGA-based solution is the best answer to the rapidly evolving deep learning industry.
While a custom ASIC can be designed to meet the needs of a computer neural network, DeePhi's co-founder and CEO, Song Yao, believes an FPGA is better suited to meet the demands of a fast-paced market. Unlike with a custom ASIC - which takes time to design and therefore has a long lead-time, a reprogrammable FPGA can be adjusted to meet the requirements of a constantly evolving deep learning network.
DeePhi is not the only company that relies on FPGA. Inspur, a leading data center product and solutions provider, is using Intel's Xeon+FPGA chip, Knight Landing, as the main piece in their data-based solutions.
If FPGA have these advantages over an ASIC, why aren't more people using them?
The biggest problem with FPGA is that it takes time to program. DeePhi's products required several months of hand coding and then debugging.
Also, there are instances where FPGA are only available on a lead-time basis. For companies that are looking to meet a certain deadline - whether's it's a time frame for their prototype or their customer's requirements - lead times are never good news.
Then there are instances where lead times stretch to two, three months. In an instance like this it makes sense to turn to the open market and rely on one company for your next FPGA purchase.
Unlike other independent distributors, EarthTron is franchised for several lines but has the ability to sell other manufacturer products as well. In this instance, EarthTron is franchised for Bomar Crystals and Transcend MicroSD cards but also deals with FPGA. We work with both Xilinx and Altera, accepting both RFQs and any excess on hand.
If you are looking for FPGA, don't forget to check out our index page. Not all FPGA are listed there so be sure to let us know if you don't find what you are looking for.