FLEX LOGIX BOOSTS AI ACCELERATOR PERFORMANCE 

                 AND LONG-TERM EFFICIENCY 


Embedded FPGA (eFPGA) can reduced memory bandwidth requirements
by more than 10x and allows efficient execution of future operators and
                                       activation functions 


MOUNTAIN VIEW, Calif. – July 2nd, 2024 – Flex Logix® Technologies, Inc., the leading
supplier of embedded FPGA (eFPGA) IP and reconfigurable DSP/SDR/AI solutions, today
announced additional applications for embedded FPGA to improve the value proposition for
AI Accelerators. 


First, memory bandwidth – the scarcest resource in AI Accelerators, especially in the cloud,
where model weights/parameters exceed 100 billion and HBM memory is expensive and
scarce. Techniques for saving memory bandwidth are evolving faster than hardware.  


“Embedded FPGA (eFPGA) can enable innovations in sub-INT4 data and weight
representations (e.g. ternary, 2 bit, 3 bit, mixed or mat-mul free) to be converted on the fly
by eFPGA into existing TPUs,” said Cheng Wang, Flex Logix CTO & SVP Software
+ Architecture. “This can also be mixed with innovations in sparsity that can further reduce
the memory bandwidth requirements. Aggregate memory bandwidth reduction can be up to
16x.” 


Second, higher performance. AI models are rapidly evolving. With most TPUs, new operators
and activation functions must be handled by a much slower processor. eFPGA can be used to
run the new operators and activation functions at much higher performance.  


Flex Logix is already using these concepts in its own InferX AI optimized IP for edge vision AI
models and DSP.   


Get more information on EFLX eFPGA and a block diagram of weight memory reduction at
www.flex-logix.com.  


About Flex Logix  

Flex Logix is a reconfigurable computing company providing leading edge eFPGA, DSP/SDR
and AI Inference solutions for semiconductor and systems companies. Flex Logix eFPGA
enables volume FPGA users to integrate the FPGA into their companion SoC, resulting in a
5-10x reduction in the cost and power of the FPGA and increasing compute density which is
critical for communications, networking, data centers, microcontrollers and others. Its
scalable DSP/SDR/AI is the most efficient, providing much higher inference throughput per
square millimeter and per watt. Flex Logix supports process nodes from 180nm to 7nm, with
5nm, 3nm and 18A in development. Flex Logix is headquartered in Mountain View, California
and has an office in Austin, Texas.


For more information, visit https://flex-logix.com.
For general information on InferX and EFLX product lines, visit our website.



Copyright 2024. All rights reserved. Flex Logix and EFLX are registered trademarks and
INFERX is a trademark of Flex Logix, Inc.