I believe this confirmation of an “active alliance” with BrainChip is new on the Tata Elxsi website?
“
Why Tata Elxsi?
(…)
- Active alliances with Brainchip for neuromorphic computing and IISc Bangalore for edge cybersecurity.”
Edge AI solutions optimize device performance with low-latency, secure AI processing. Explore its impact on industries like healthcare, retail, and electronics.
www.tataelxsi.com
View attachment 88392
View attachment 88393
View attachment 88394
View attachment 88395
View attachment 88398
View attachment 88399
Despite our years of collaboration with TCS researchers and the above encouraging affirmation of an “active alliance” with Tata Elxsi, we should not ignore that TCS are also exploring other neuromorphic options for what will ultimately be their own clients.
And while a number of TCS patents do refer to Akida as an example of a neuromorphic processor that could be utilised, they always
also refer to Loihi, as far as I’m aware of.
A recent case in point for Tata Consultancy Research’s polyamory is the June 2025 paper “The Promise of Spiking Neural Networks for Ubiquitous Computing: A Survey and New Perspectives”, co-authored by five Singapore Management University (SMU) researchers
as well as Sounak Dey and Arpan Pal from TCS, both very familiar names to regular readers of this forum.
Although we know those two TCS researchers to be fans of Akida, they sadly did not express a preference for BrainChip’s neuromorphic processor over those from our competitors in below paper published less than six weeks ago.
On the contrary, in their concluding “key takeaway” recommendations of neuromorphic hardware (“We make the following recommendations for readers with different needs considering neuromorphic hardware chipsets”), the seven co-authors do not even mention Akida at all.
Even more surprisingly, the section on Akida is factually incorrect:
- AKD1500 is a first generation reference chip and is
not based on Akida 2.0, BrainChip’s second generation platform that supports TENNs and vision transformers.
- An AKD2000 reference chip does not (yet) exist - it may or may not materialise. At present, only Akida 2.0 IP is commercially available - not an actual silicon chip, as claimed by the paper’s authors.
- The paper is in total ignorance of ultra-low-power Akida Pico, operating on less than 1mW of power, which was revealed by our company back in October 2024 and is based on Akida 2.0.
It is highly unlikely this (possibly revised) version of the paper published on 1 June 2025 would have been submitted to arXiv prior to BrainChip’s announcement of Akida Pico, and we can safely assume Sounak Dey and Arpan Pal would have been aware of that October 2024 BrainChip announcement (unlike maybe their SMU co-authors).
One could argue the reason Akida Pico is not mentioned could possibly be that an actual Akida Pico
chip is not commercially available, yet, given the authors state
“5.2 Neuromorphic Hardware
In this subsection,
we summarize the latest commercially available neuromorphic hardware chipsets, highlighting their capabilities and development support for building and deploying spiking neural networks.”,
which, however, in turn begs the question, why Loihi 2 is listed, then, as it was always conceptualised as a research chip and is not commercially available. In the paragraph on Loihi 2, the authors correctly state that “this neuromorphic research chipset is available only through the Intel Neuromorphic Research Community (INRC).”
Given the fact that Sounak Dey and Arpan Pal co-authored this paper, the above inaccuracies are bewildering, to say the least. Did the two TCS researchers who both have firsthand experience with Akida contribute to only part of this paper and not proofread the final version before it was submitted?
Either way not a good look…
(…)
(…)