https://research.ibm.com/blog/northpole-ibm-ai-chiphttps://brainchip.com/akida-generations/Everyone can decide themselves which platform sounds more like the human brain
.
"for Modha, this is just one important milestone along a continuum that has dominated the last 19 years of his professional career. He’s been working on digital brain-inspired chips throughout that time, knowing that the brain is the most energy-efficient processor we know, and
searching for ways to replicate that digitally. TrueNorth was fully inspired by the structures of neurons in the brain — and had as many digital “synapses” in it as the brain of a bee. But sitting on a park bench in 2015 in San Francisco, Modha said he was thinking through his work to date. He had the belief that there was something in marrying the best of traditional processing devices with the structure of processing in the brain, where memory and processing are interspersed throughout the brain. The answer was “brain-inspired computing, with silicon speed,” according to Modha."
Wow, 20 years of research in neuromorphic computing with billions of dollars invested and IBM has failed to bring TrueNorth to fruition and failed to release a commercially viable digital neuromorphic edge AI platform
.
So what does IBM do to show that they have not wasted time and money?...NorthPole..."Over the next eight years, Modha and his colleagues were single-minded and hermetic in their goal of turning this vision into a reality. Toiling inconspicuously in Almaden, the team didn’t give any lectures or publish any papers on their work, until this year. Each person brought different skills and perspective yet everyone collaborated so that as a whole the team’s contribution was much greater than the sum of the parts. Now, the plan is to show what NorthPole could do, while exploring how to translate the designs into smaller chip production processes and further exploring the architectural possibilities."...LOL, it's really a smaller "commercial" version of TrueNorth.
This article in 2014,
https://spectrum.ieee.org/how-ibm-got-brainlike-efficiency-from-the-truenorth-chip, gives an idea of the pressure this Modha guy is feeling so he has to at least come up with something. "The chip (TrueNorth) represents the culmination of a decade of Modha’s personal research and almost six years of funding from the U.S. Defense Advanced Research Projects Agency (DARPA). Modha continues to lead DARPA’s SyNAPSE project, a global effort that has committed more than US $100 million since 2008 to making computers that can learn. “Our long-term end goal is to build a ‘brain in a box’ with 100 billion synapses consuming 1 kilowatt of power,” Modha says."
So what did the "geniuses" at IBM really came up with?
TrueNorth has 5.4 billions transistors in 4096 cores to give it 1 million programmable neurons and 256 million programmable synapses. NorthPole has 22 billions transistors in 256 cores. If you bother reading about TrueNorth, it's actually the number of cores that determines the number of synapses, so NorthPole have simply increased the number of transistors to compensate for less cores so that they can squeeze it into a 12nm process. That means NorthPole has potentially 5 million programmable neurons but only as little as 16 million programmable synapses...if I'm generous I would estimate between 16 to 100 million synapses. What about Akida 2.0?...
1.2 million neurons and 100 billion synapses in 256 nodes ...in a 20nm process but if required can be easily be scaled down to 12nm. Remember an elephant has 3x the amount of neurons than humans but only a fraction of synapses, plus their neurons are in the wrong places.
It's obvious that IBM has it's US roots sewn in so many facets of american society that such a mention of a neuromorphic chip has everyone hyped up.It seems that the tech world has just woken up to what Brainchip has already done 3 years ago so I guess Brainchip is truly about 3 years ahead.
Take this article,
https://defensescoop.com/2023/10/19/what-ibms-new-ai-enabling-brain-inspired-computer-chips-could-mean-for-dod/This Holliday guy must have suddenly woken up from sleep or a close friend of this Modha dude, "One of the government’s top engineering and robotics experts and technology innovation leaders, Holliday has helped push forward a range of crucial national security-aligned capabilities. “This always blows my mind when I think of it. So, our brains have 10 to the 14th or 100 trillion synapses, and we have [billions of] neurons. Humans have created civilization, and our brain consumes the power of a light bulb, right? And it occupies the volume of a large soda bottle. So, you know, it’s amazing,” he said. Modha “is right about this being a slight reflection of the brain,” Holliday added, in that the architectures are in many ways similar. However, the NorthPole prototype chip currently has 22 billion transistors — compared to the billions or trillions of synapses in human brains."......
WTF Please someone introduce this guy to Akida 2.0
As I have said many times before, this
ratio of neurons to synapse is the key to intelligence. The human has a ratio of 1:10,000 and Akida has 1:100,000 which is potentially "smarter" but there is no way 100 billion neurons can be squeezed in one chip...I bet you PVDM is stacking thousands of Akida in his private lab!
Brainchip holds the patents on how to create synapses between digital neurons. It's taken IBM over 20 years and it seems they still can't form anywhere close to the amount of synapses between these groups of "transistors" that acts as programmable neurons. Now they just have to go with what they have and just try to crank up the amount of "transistors"...to the lay person, it sounds fantastic
It's the amount of synapses that allow Akida to have TENNS (spatio-temporal processing), inbuilt ViT, multi sensory simultaneous processing, skip connections, at memory compute, real-time event-based processing, continuous on chip one-shot learning and flexible scalable platform to accept any previously trained format.
"The researchers found that NorthPole can recognize images faster and is 25 times more energy efficient when it comes to the number of frames interpreted per joule of power required, than common 12-nm GPUs and 14-nm CPUs like those now widely used and made by Nvidia and other major players." It seems these "genuises" from IBM are getting off on NorthPole being faster and more energy efficient than the CPU/GPU...LOL, Akida has long past this phase of proving that!
"Modha has been working to generate the prototype that NorthPole has become since the early 2000s for IBM, with support from the Defense Advanced Research Projects Agency, the Air Force and others along the way.When Holliday returned to the Pentagon in 2021 to serve in his latest post, he and his team were approached about funding IBM’s NorthPole pursuit. “Unfortunately, it had been canceled by the previous administration,” he told DefenseScoop. Advertisement“They came to me to say, ‘We are close. This is a capability we absolutely need — because it provides AI compute at the edge,’” Holliday recalled. The Office of the Secretary of Defense then opted to fund them to finish the job so that the chips could be fabricated in low numbers in the U.S. and be tested and evaluated by the government and the industrial base.Now, that’s happening across federal, defense and other research organizations.Each prototype AI chip has to be placed on a field programmable array to then be integrated to work within more complex systems. Looking to ultimately operationalize the prototype, Holliday confirmed that IBM hosted a transition workshop this summer to teach government lab insiders and other partners how to program those arrays so that they can puzzle out and run new use cases on the chips.“IBM has proved that this is a leap-ahead capability. And we need to bridge one of these ‘valleys of death,’ which is getting it out of the prototype stage, and doing test and evaluation, and after that point, giving feedback to IBM and having them iterate one or two more times. And then, this goes into volume production for platforms that support both commercial, as well as for DOD applications,” Holliday said of the vision. Advertisement NorthPole has been funded most recently via OSD’s microelectronics program. For the next iteration, he’s encouraging IBM to go after investments via the CHIPS and Science Act.“They’ve been in the lab for almost two decades and it’s ready to be ‘fab’ now — and so that’s one of the things that the CHIPS Act was designed to address is to get capabilities like this to commercial scale. So they fab it in low quantities at Global Foundries in New York, but to get to commercial scale they need to get it to a commercial-scale [fabrication facility]. There are some in the U.S. but not at the state-of-the-art technology nodes. And so TSMC and Intel have announced, and actually broken ground on three different fabs … but it’s going be years before they’re online,” Holliday explained. Modha confirmed that — beyond the exploration of next steps, even smaller architectures, and new directions for this research — his team has already “started the process of redesigning the circuit board, so as to be less vulnerable to some of the supply chain constraints.”"
This says it all about the state of play for Brainchip. The guys at IBM and the guys that control the direction of US Defense are all waiting to dip their hands together in the pie, ignoring the fact that there is a more superior pure digital, pure neuromorphic platform that is already taped out with both TSMC and Global Foundries and proven on ARM's most sophisticated AI silicon platform. I'm not sure what else Brainchip's management can do. Maybe they wish to NOT to go down this pure US direction and that's why they are "playing" around with the Japanese, Indian, European and Isreali instead
, knowing that the US government can be very restrictive.
Thinking about this, I think with IBM (and Intel) pushing the tech world towards neuromorphic computing, it can be the catalyst Brainchip needs to make a run. If I can decipher the BS that is happening then I'm sure there are many others that can also see how amazing Akida is...just wait until PVDM and Anil open up the synapses to create more "at chip memory", something the genuises at IBM are getting off on. Like our brain where we only use 10%, Akida has not gotten close to using a fraction of their neurons and synapses. Again it's all about synapses and turning up the memory is all about opening up synaptic channels. Personally I think PVDM and Anil already know how to do this but they are keeping this up their sleeves, especially realizing how slow and deliberately resistant the tech world can be. When news like NorthPole comes out, I'm actually excited that we are able to progress further towards a neuromorphic computing world and Akida will become more sophisticated.
(20min delay)
