Skip navigation

FPGA as a multitenancy platform

5G technology has enabled communication bandwidths never achieved before, but, at the same time, it imposes huge pressure on processing architectures to ingest the data at matching rates. The task for Shaji, a summer intern in the Cloud Systems and Platforms department at Ericsson Research, was to contribute to the ongoing work of enabling FPGAs (field-programmable gate arrays) as high-performance multitenancy accelerator platforms.
Machine room for multitenancy platform

Internship focus

FPGAs are known for parallelism and reconfiguration, but, lately, the dynamic partial reconfiguration feature is gaining traction for different applications, such as on-demand accelerator deployment, mitigating radiation induced faults, adaptive computing, and more. Amazon, for example, is now offering the EC2 F1 instance in which the user can deploy custom-made accelerators on the provided FPGA.

With the advent of 3D ICs based on stacked silicon interconnect (SSI) technology, FPGA manufacturers are offering devices that are very rich in resources, and the accelerator designs implemented on these devices usually occupy 30-40% of the resources. Dedicating one large FPGA to one user increases the infrastructure cost greatly. So, based on these facts, the main idea is to share the FPGA among different users by introducing virtualization based on dynamic partial reconfiguration capability.

Internship goal

To enable virtualization, an FPGA is divided into two main regions: static and dynamic. The static region holds the design that provides the infrastructure to load and control designs into the dynamic region. The dynamic region is intended to be split further into subregions to be dedicated to different users. There is a lot of ongoing research regarding the problems associated with FPGA virtualization and multitenancy, with the main challenges designated to the author including:

  1. proposing interface layers between the static and dynamic subregions
  2. sharing and regulating the limited PCIe bandwidth across the different users with minimum overhead in terms of resources and throughput
  3. ensuring data security for different subregions

My experience overall

Working at Ericsson Research was an excellent opportunity to exercise my academic knowledge and put it to use for the realization of a state-of-the-art research idea. Besides improving my personal skills, I also got to make a lot of new friends and build up a wider professional network. It feels to me like my internship period passed in the blink of an eye, and I am so happy to have had the chance to take part in this opportunity. I would like to conclude by quoting something I really like which was said by an Ericsson employee on the Summer Interns Introduction Day:

“If you are not sad, embarrassed, or in love, you are not taking enough risks.”

About the author

Shaji Farooq Baig - Ericsson Intern Blog

Shaji Farooq Baig comes from Pakistan and is about to complete his master’s degree in embedded systems at KTH Royal Institute of Technology. He is a technology enthusiast and has research interests in reconfigurable architectures and high-performance computing. Helping people is one of his main pursuits, and he has been an International Student Ambassador at KTH for two years with the main task of responding to the queries of prospective and newly admitted students.