Menta Predictions – Disrupting Adaptive Compute at The Edge

Yoan Dupret

Last year, we talked about the fracturing of the old computing order and how no one can predict anymore the winning architectural or algorithmic solution for a given compute or communication problem – even within a two year event horizon. We used AI as a case in point – which turned out to be pretty accurate as we saw the emergence of companies with brand new architectures, and all very different, such as AnotherBrain, Blaize, Hailo or SiMA.ai, while some others already disappeared or pivoted their products.

We also talked about how sensors are becoming ubiquitous in our lives, the trend to customize merchant chips and the tendency toward highly heterogenous SoC architectures.

We concluded that this not only means more chips but also more changes in algorithms, architectures and interconnects which would lead to the increasing utilization of field reconfigurable device, such as the ones enabled by Menta embedded FPGA IP – in short ‘adaptive compute’ devices. This also turned out to be accurate.

What we did not predict though was the global sanity crisis of Covid-19. Despite this, several eFPGA IP providers announced design wins in 2020 and at Menta we did particularly well as we doubled our revenues – though we would have done much better without the virus.

2021 will see the acceleration of the trends we described last year as demonstrated by the recent M&A announcements – especially the acquisition of ARM by nVidia and Xilinx by AMD.

On one hand, nVidia, the leading provider of accelerators in the form of GPUs is buying a general-purpose CPU IP provider and reportedly wants to migrate its compute GPU IP to Edge devices through ARM. This is a great endorsement of the IP business model where ARM is the world leader.

On the other hand, AMD, who is both a CPU and a GPU provider, is adding additional acceleration functions in the form of programmable logic to its portfolio through the Xilinx acquisition.

This trend to add programmable logic accelerators, in the cloud and data center world, started with Microsoft Catapult project, disclosed in 2014, went to the next level when Intel acquired Altera for $16.7B in 2015 and naturally reached a climax with the $35B proposed acquisition of Xilinx by AMD in 2020.

All this is about adaptive compute acceleration beyond the CPU and shows that FPGA technology is essential for getting there. After all AMD has both the CPU and the GPU yet still needed the FPGA.

Now, this movement that started within the Cloud and data-centers, cannot stop there. Computation cannot happen solely in the Cloud as transporting terabytes of data from the Edge to the Cloud makes no economic or technical sense. The number of chips for edge devices will therefore keep growing in 2021 – with the same requirements for adaptability and programmability as the Cloud but with huge constraints on power consumption, size and, most of all, cost.

We will therefore see a strong increase in the requirements for embedded programmable logic for high volume SoC Edge devices – with the demand at least doubling in 2021. We believe the embedded FPGA approach is the only proven option for adaptive compute at the Edge.

At the same time, 2021 will see more designs than just AMD Ryzen making use of the chiplet approach as a way of disaggregating complex chips into smaller physical units. We could also probably see the first AMD product integrating a Xilinx FPGA in the form of a chiplet potentially in 2021.

Chiplets are now a clear path toward cost, power and time to market reductions for data-center devices. The natural next stage is to embrace the chiplet approach at the Edge. There are still some hurdles to cross, but we expect the first solutions to start emerging during 2021. Such an approach will strongly facilitate the use of embedded FPGAs within chiplet based designs.
Stay tuned as Menta will unveil a number of innovations during 2021 that will revolutionize chiplet usage at the Edge.

About Author

Yoan is the Managing Director and VP of Business Development at Menta – a leader in embedded FPGA IP cores for chips and smart sensors. Prior to his position at Menta, he held various managerial and technical positions at DelfMEMS, Samsung, CSR, Infineon and Altis Semiconductor. Yoan holds a PhD from Supelec (France) and an Engineering degree (MSEE) from ESEO (France).

Tags: Predictions

This entry was posted on Tuesday, February 2nd, 2021 at 4:37 pm. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.