https://arxiv.org/pdf/2205.13037.pdfThis is a great article...technical but serious long term investors should at least try to understand why neuromorphic computing will be the next big wave of technology to support the rise of AI...it's no longer a question of "if", it's more of "when"...it intrigues me how people can imagine AI without having an artificial brain that function like a human brain i.e, neuromorphic computing. The following are the important features of neuromorphic computing and why AI has to exist in a neuromorphic platform:
• neural network - based approach: the capability of learning (no need to set the parameters explicitly), the emergence of intellectual properties by linking a large number of relatively simple elements into a network,
• parallelism: parallel work of neurons, simultaneous execution of different tasks,
• asynchrony: no single synchronizing process,
• impulse nature of information transmission: minimum overhead for signal transmission and signal processing at the receiving neuron, resistance to noise,
• local learning: lower overhead for data transfer operations during learning, the ability to create unlimitedly large systems, continuous and incremental learning,
• sparsity of data streams: event-driven signal processing, lower overhead for data transfer and data processing,
• analog computing: efficient hardware implementation.
• in-memory computing: no overhead for transferring intermediate data, no competitive memory access.
There is certainly not a long list of pure neuromorphic players and an even lesser list of current commercially available platforms. Besides IBM and Intel, I don't see other big tech on this list...this means that a neuromorphic platform that can be licensed as an IP for any big tech to build their custom solutions will likely be the most ubiquitous neuromorphic platform...the Renesas/ARM/Akida combination chip will get this ball rolling by the year's end.
To me the progress and struggle of Brainchip to develop and evolve the Akida neuromophic SNN is no different than when NVIDIA started to develop and evolve their GPU...how I wish to have invested in NVIDIA at the beginning of their journey
...Akida's journey is a reflection of history repeating itself in just a different way!
To me, the 1st generation Akida was a necessary stepping stone in the evolution of the Akida "brain"...the current human would certainly view our ancestor's homo sapien brain as "useless" but we would not be developing AI now without having the primitive brain!
Anyway the 1st generation Akida enabled the development of peripheral supporting "structures" eg, Meta TF, integration with ARM, taping out with TSMC and Global Foundries, integration with algorithm development platforms such as Edge Impulse etc..
Now with further generation Akida, new gates of "intelligence" will be opened without drastically changing the core neuromorphic structure i.e, TENNS with 2nd generation Akida...which is a feat that has been underestimated by the market but not by the engineers developing the next great edge AI product...solving the the riddle of both spatial and temporal processing in real time with event based data at the sensor itself is something unimaginable in the past!
There's some here saying that they have sold out in recent days...well, I'm one of those happy to slowly add more to my sizeable collection
(20min delay)