Design & Reuse

Industry Leading PPA EFPGA And DSP/AI In Development For TSMC N5/4/3

Improved EFLX® eFPGA enables Millions of LUTs and superior DSP/SDR/AI

flex-logix.com/, Apr. 08, 2024 – 

MOUNTAIN VIEW, Calif. – April 9th, 2024 – Flex Logix® Technologies, Inc., the leading supplier of embedded FPGA (eFPGA) IP and reconfigurable DSP/AI solutions, announced that it is in development of next generation EFLX eFPGA Generation 3.0 for TSMC N5/4 and N3. EFLX Generation 3.0, along with the eXpresoTM eFPGA compiler enables Millions of LUTs and higher frequency operation; and LUTs that can programmably be I/O's for applications requiring a lot of control like networking switches or InferX DSP/SDR/AI. EFLX Gen3.0 can reconfigure in a few microseconds for dynamic workload switches.

EFLX Gen 3.0 is the control hub for industry leading InferXTM DSP/SDR and AI PPA (performance/power/area):

  • just 4mm2 of N5 InferX silicon can deliver 8.5Gigasamples/second of Complex INT16 dynamically switchable 1K/2K/4K FFT; or 234 Inferences/second (IPS) for YOLOv5s or 33 IPS for Megapixel DETR 2020 Transformer model.
  • Just 25mm2 of N5 InferX/EFLX silicon can implement a reconfigurable 64 Gsamples/second OFDM Software Defined Radio (SDR). See details at https://flex- logix.com/inferx-dsp.
  • InferX scales up linearly for more throughput as needed.

"Flex Logix is the leader in embedded FPGA IP with more than 20 licensed customers, more than 40 licensed chips and over 25 chips in silicon," said Geoff Tate, CEO of Flex Logix. "Our commercial customers rely on our TSMC EFLX eFPGA technology already from 40nm to 7nm for applications including reconfigurable crypto agility, micro FPGA, programmable GPIO, customer-programmable accelerators, reconfigurable packet processing, asynchronous control and many more. We are now supporting our customers on TSMC N5/4/3 with our improved EFLX architecture and our new InferX DSP/AI hardware and the InferX compiler. Customers can use our DSP/SDR/AI IP to outperform the fastest FPGAs for signal processing and SDR and the fastest edge vision GPUs on Inference AI."

click here to read more...