Easing the Effort: Mipsology Accelerates ML with Zebra FPGA IP

June 09, 2021 by Sam Holland

Today, Mipsology is announcing its solution for FPGA programming: Zebra FPGA IP. This IP aims to ease FPGA programming, giving the designer more time to design.

To add to today's earlier announcement from Xilinx on its Versal AI Edge family, Mipsology, a startup focused on the acceleration of deep learning inferences, also has something to talk about.

Today, Mipsology is introducing its Zebra FPGA IP (field-programmable gate array intellectual property). This FPGA IP is designed to expedite the development and deployment of both FPGA and SoC-based machine learning systems, hoping to ease the requirements for programming FPGAs by removing the need for a CPU or GPU.


Replacing the GPU to Ease FPGA Programming 

The Zebra FPGA IP is a neural network acceleration solution that is based on the startup's flagship software, Zebra. Since FPGA programming can often be difficult, the technology aims to assist designers who develop machine learning-based systems in the edge and embedded computing market, making it simpler to program and integrate.


The inference between GPU vs. Zebra FPGA IP.

The inference between GPU vs. Zebra FPGA IP. Image used courtesy of Mipsology


With the increasing need for edge computing in neural networks and embedded computing, Mipsology's FPGA IP adaptability can be crucial. One reason is that application-specific integrated circuits (ASICs) and GPUs aren't as well-equipped to keep up with the fast evolution of neural networks. 


An example of a neural network overview. Image used courtesy of Aldec


Mipsology talks about how, through combining the Zebra FPGA IP and the ability for developers to program FPGA's hardware, it can provide a malleable ML platform for systems at the edge. 

The company also claims that the IP can enable designers to fully utilize Arm CPUs and GPUs since the Zebra FPGA IP requires developers to use only part of the FPGA. This utilization, as Mipsology claims, leaves users with "the highest possible density of computation" to work with, resulting in the ability to deliver functions with both high-performance and low latency throughout the development process.


Recommended Applications

Mipsology acts on developers' increasing need for AI acceleration through the Zebra FPGA IP for both edge and embedded computing applications. Accordingly, the IP is optimized to power a wide range of edge applications like:

  • industrial automation
  • security
  • autonomous vehicles
  • smart cities

The technology could be suited for power AI-based camera systems, particularly owing to Zebra FPGA IP's ability to be paired with video decoders and Arm CPUs. In addition to the IP, this ability, being easily compatible through plug and play with cameras, set-top boxes, mobile systems, and more, allows Mipsology's FPGA technology to become a possibly efficient, long-lasting solution over GPUs.


Easier to Evolve

The focus on AI development is growing faster than ever before. Developments that make it easier to integrate, as well as remove potentially nonessential components are in constant demand. Mipsology's solution attempts to give the designers more time to evolve instead of losing time to other tasks.