BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
I like your thinking P...
It became apparent to me at least when I asked the question directly to an unnamed person with regards the taping out of AKD II, having read that both Todd and Anil had mentioned it had commenced, well, a no response sealed the deal for me.
I have been monitoring the patents sites weekly, some of which * 2017* it appears the examiner is playing ping pong, or in a coma !
As you mentioned, nice to hear the odd bit of new info from Tappsie surrounding positive developments with AKD II
DingoBorat
Slim
What have I missed Diogenese?..
What's this about a "thinking" neuron and AKIDA 2.0 IP still being finalised now?
It was announced around 2 years ago now, that AKIDA 2.0 IP was available and now no announcement that they are still tinkering?..
Just an EETimes podcast thingo?
Isn't that something requiring an ASX release/correction, if something previously released on the ASX was incorrect?
Last edited:
uiux
Regular
Below is a consolidated analysis of the BrainChip podcast—from both a technical perspective (with emphasis on neuromorphic technology, edge processing, and GPU contrasts) and a market/ASX stock analyst viewpoint.
1. Technical Overview
Neuromorphic & Temporal Event-Based Neural Networks (TENN)
- Core Innovation:
BrainChip’s primary technical advance centers on their Temporal Event-Based Neural Networks (TENN). Unlike traditional feed-forward or standard recurrent networks (such as LSTMs), TENNs are designed to leverage long-range temporal dependencies with an internal state that continuously “remembers” past inputs. This leads to networks that are both compact (fewer weights) and capable of processing time-varying signals in real time. - Physical Inductive Bias:
By incorporating physical constraints (using tools such as Legendre polynomials), TENNs encode real-world continuity—for example, ensuring that an object’s movement adheres to realistic acceleration or jerk limits. This physical inductive bias not only reduces the volume of training data required but also improves training speed and accuracy relative to conventional convolutional approaches that lack such constraints. - Training & Inference Trade-Offs:
BrainChip’s approach cleverly marries the best of both worlds:- Training Efficiency:
By initially training in a convolutional-like (parallelizable) domain, they exploit the strengths of GPU processing for fast convergence. - Inference Efficiency:
The trained model is then transformed into a recurrent form that dramatically reduces memory footprint and energy consumption—critical for edge devices.
- Training Efficiency:
- Event-Based Processing & Sparsity:
BrainChip’s hardware (e.g., the Akida chip) is built around event-based processing. In contrast to conventional spike-based models (where a “spike” is a binary event), their approach allows for multi-bit “events” that carry amplitude and richer information. This, combined with inherent sparsity (only non-zero activations trigger computation), offers:- Reduced Computational Overhead:
Lower power consumption and faster execution because only active events are processed. - Parallelism at the Event Level:
While traditional GPU pipelines excel at fixed-size parallel operations, BrainChip’s design adjusts dynamically to sparse data flows—a key advantage for variable-rate, edge data such as sensor streams, audio, or video.
- Reduced Computational Overhead:
- Comparison with Competing Models:
The podcast contrasts TENNs with parallel research (e.g., the Mamba model):- TENNs:
Feature a compact internal state suited for low-power, edge inference. - Mamba:
Uses a large internal state bank optimized for GPU-based inference (useful in large language modeling).
The key insight is that while both architectures are promising, TENNs are more naturally aligned with the constraints of on-device (edge) computing.
- TENNs:
Hardware Evolution – The Akita 2.0 Architecture
- Multi-Resolution & Flexibility:
The next-generation Akida chip is set to offer multiple operating modes—from single-bit (spike-like) to 4-bit and 8-bit resolutions, with future plans for 16-bit. This flexibility is critical as it allows the chip to balance between the energy-efficient processing required at the edge and the computational demands of more data-intensive applications. - Integration & Programmability:
While the hardware is fundamentally an event-driven state machine (echoing the hardwired nature of biological systems), some degree of programmability is built in. This ensures that the chip can adapt to a variety of AI workloads without being locked into a single algorithmic framework.
2. Market & ASX Stock Analyst Perspective
Business Model & Strategic Positioning
- IP Licensing Focus:
BrainChip’s primary revenue model is based on licensing its IP rather than mass-producing chips. This allows them to partner with established semiconductor manufacturers and integrate into existing supply chains—a less capital-intensive route to revenue. - Edge AI Opportunity:
With the growth in IoT, autonomous systems, wearable health monitors, and real-time applications (like speech-to-text and gesture recognition), the demand for low-power, efficient processing at the edge is surging. BrainChip’s neuromorphic approach is well positioned to capture a significant share of this emerging market niche. - Co-Evolution of Hardware and Algorithms:
The close integration between their algorithmic research (TENN and related architectures) and hardware development (Akida, Akida 2.0) creates a strong moat. By continuously iterating on both fronts, BrainChip can offer a more optimized solution than competitors who focus on either software or hardware in isolation.
Competitive Landscape & Differentiators
- Edge vs. GPU-Centric Models:
As GPUs remain dominant in large-scale training but struggle with power efficiency for real-time edge applications, BrainChip’s technology provides a complementary solution. Their IP could be increasingly attractive to OEMs looking to offload AI inference from power-hungry GPUs. - Sparsity & Event-Driven Processing:
The ability to capitalize on sparsity—processing only active events—means that BrainChip’s solutions can achieve significant energy savings, a key selling point in sectors such as automotive, industrial monitoring, and biomedical devices. - Market Timing & Partnerships:
Being a relatively mature player (established in 2004) gives BrainChip credibility, but the pivot toward commercialization and IP licensing is a critical inflection point. Success in forming key partnerships (with major semiconductor firms or OEMs) will be a major driver for revenue growth.
Investment Considerations for ASX Investors
- Growth Potential:
The global push for edge computing and low-power AI solutions offers a significant long-term market opportunity. If BrainChip’s technology is widely adopted, it could lead to robust licensing revenues and a strong competitive position. - Risk Factors:
- Technology Adoption:
Neuromorphic computing is still in its relative infancy compared to conventional GPU/CPU architectures. Market acceptance may take time. - Competitive Pressure:
Major players (e.g., NVIDIA, Intel) and other specialized startups are also exploring low-power AI solutions. BrainChip must maintain its technological lead and secure strategic partnerships. - Ecosystem Integration:
As noted by BrainChip’s CTO, the chip must fit within broader software and hardware ecosystems. Failure to integrate seamlessly could impede market penetration.
- Technology Adoption:
- Valuation & Catalyst Outlook:
For ASX investors, the company’s stock could represent a high-growth, high-innovation opportunity if it can demonstrate both technical leadership and successful commercialization. Key upcoming catalysts might include:- Announcements of major licensing agreements or partnerships.
- Successful deployment of Akifa 2.0 in real-world applications (e.g., automotive, wearables).
- Third-party validation (competitive benchmarks, academic endorsements) of TENNs performance versus traditional models.
3. Conclusion
Technical Perspective:
BrainChip’s approach—integrating a novel recurrent neural architecture (TENN) with a flexible, multi-bit, event-driven chip design—positions it at the frontier of neuromorphic and edge computing. Their solution addresses the limitations of both traditional convolutional networks and GPU-optimized recurrent networks by blending efficient training paradigms with a low-power, real-time inference engine.
Market/Investment Perspective:
From an ASX stock analyst’s viewpoint, BrainChip is innovating in a high-growth niche. Their IP licensing model, combined with a robust technology platform and a focus on edge applications, makes them a compelling long-term play. However, investors should remain cautious about execution risks, ecosystem integration challenges, and competitive pressures.
Overall, if BrainChip continues to validate its technology with strong performance metrics and strategic partnerships, it could be well-positioned to capitalize on the accelerating demand for edge AI solutions—potentially offering significant upside to shareholders.
This analysis reflects the interplay between cutting-edge neuromorphic design and the evolving market for energy-efficient, real-time AI processing, making BrainChip a noteworthy candidate for both technical and investment scrutiny.
The podcast indicates that BrainChip isn’t in a research phase anymore—they’ve already shifted toward commercialization by focusing on IP licensing. In fact:
- Strategic Pivot:
Tony Lewis explained that about two to three years ago the company transitioned from pure research into a commercialization model. Their goal became to package their neuromorphic designs (like their neural processing engine for TENNs) as IP that other chip makers can license. - Proof-of-Concept Chips:
Although they occasionally fabricate their own chips (like the Akida series) to validate and demonstrate the performance of their technology, the primary revenue driver is to license this IP rather than mass-produce chips themselves. - Current Position:
The discussion makes it clear that BrainChip’s business model is built around IP licensing for edge AI applications. This suggests that they are already well on their way—or even actively engaged—in pursuing licensing agreements, rather than being far away from this stage.
The overall tone of the interview is decidedly positive. Key points that support this include:
- Optimism About the Technology:
The speakers consistently express enthusiasm and confidence in BrainChip’s neuromorphic approach, particularly with their TENNs architecture. They highlight significant breakthroughs—such as efficient training, low-power edge inference, and strong performance across various applications (e.g., eye tracking, ASR, and gesture recognition). - Business Model Confidence:
The discussion on shifting toward an IP licensing strategy is presented as a mature, deliberate pivot rather than a fallback. BrainChip is portrayed as leveraging its technological advances to build a commercially viable model, emphasizing both the research accomplishments and the practical steps taken toward market penetration. - Competitive Positioning:
The interview underscores BrainChip’s competitive advantage—especially in edge computing—by comparing their approach favorably against traditional methods and other emerging models (like Mamba). This comparative optimism suggests that the company believes its technology not only stands out but is also ahead of many current alternatives. - Future Prospects:
The speakers are forward-looking, discussing upcoming iterations (such as Akida 2.0 and even a planned 16-bit architecture) and potential market applications. This forward momentum indicates a positive outlook for both technological and commercial growth.
SharesForBrekky
Regular
Bravo
If ARM was an arm, BRN would be its biceps💪!
It'll be very interesting to see if we partner up with Anduril in the not too distant future.
The reason why I say this is because I just did a little bit of poking around and discovered that Dr. Chaffra Affouda (see below), who is currently a Business Development Director at Anduril, worked as Senior Principal Systems Engineer at Raytheon Intelligence & Space from Apr 2019 to Sep 2022 before joinnig Anduril.
While he was at Raytheon, Dr. Affouda worked on developing multi-functional, neuromorphic infrared imaging sensors and cameras!
Prior to that he was with the US Naval Research Laboratory.
An employee demonstrates some of Anduril’s A.I. technology, in 2021.Credit...Philip Cheung for The New York Times
By Erin Griffith and Cade Metz
Reporting from San Francisco
Feb. 7, 2025
Sign up for the On Tech newsletter. Get our best tech reporting from the week. Get it sent to your inbox.
Anduril, an artificial intelligence military start-up, is set to complete a new round of funding that would double the value of the company to $28 billion, according to four people familiar with the negotiations.
The funding round, which is led by Founders Fund and has not yet closed, is raising up to $2.5 billion, the people said. Founders Fund alone plans to invest $1 billion, the largest check ever written by the firm, two of the people said.
Anduril designs and builds autonomous systems and weapons for the military and other government agencies, including flying drones, missiles, underwater vessels and surveillance equipment for monitoring both national borders and the battlefield. It is one of a new wave of companies building systems based on A.I. technologies for the government.
Founders Fund, started by the entrepreneur and investor Peter Thiel, has backed Anduril since its start in 2017, and one of Anduril’s co-founders, Trae Stephens, is a partner at the firm. Mr. Thiel, who also co-founded Palantir, a military technology company, has long been a backer of Republican candidates, including President Trump in 2016 and JD Vance’s run for Senate in 2022.
Six months ago, Anduril raised $1.5 billion at a $14 billion valuation.
Founders Fund declined to comment. CNBC first reported details of the funding talks.
The latest influx of cash comes as defense technology start-ups are ebullient about their prospects. Enthusiasm for building technology for the U.S. military has grown in recent years in Silicon Valley, a reversal from more than a decade of shying away from those contracts. As recently as 2018, thousands of employees at Google signed a letter protesting the company’s military contracts.
That resistance has slowly shifted, as more venture capital firms pour money into the sector.
Mr. Trump is expected to turbocharge investments further. Palmer Luckey, Anduril’s founder, has supported the president since his 2016 campaign. He donated to Mr. Trump’s campaigns in the 2016, 2020 and 2024 elections, and has hosted fund-raisers.
On the night of the presidential election in November, Mr. Luckey posted a meme celebrating Mr. Trump’s victory. Elon Musk, the tech executive and a close adviser to the president, responded, saying it was “very important to open DoD/Intel to entrepreneurial companies like yours.”
photonicsproject.org
EXTRACT FROM THE INTERVIEW
www.linkedin.com
The reason why I say this is because I just did a little bit of poking around and discovered that Dr. Chaffra Affouda (see below), who is currently a Business Development Director at Anduril, worked as Senior Principal Systems Engineer at Raytheon Intelligence & Space from Apr 2019 to Sep 2022 before joinnig Anduril.
While he was at Raytheon, Dr. Affouda worked on developing multi-functional, neuromorphic infrared imaging sensors and cameras!
Prior to that he was with the US Naval Research Laboratory.
A.I. Military Start-Up Anduril Close to Deal That Would Value It at $28 Billion
The Southern California start-up, which builds flying drones and missiles, is set to raise up to $2.5 billion.
An employee demonstrates some of Anduril’s A.I. technology, in 2021.Credit...Philip Cheung for The New York Times
By Erin Griffith and Cade Metz
Reporting from San Francisco
Feb. 7, 2025
Sign up for the On Tech newsletter. Get our best tech reporting from the week. Get it sent to your inbox.
Anduril, an artificial intelligence military start-up, is set to complete a new round of funding that would double the value of the company to $28 billion, according to four people familiar with the negotiations.
The funding round, which is led by Founders Fund and has not yet closed, is raising up to $2.5 billion, the people said. Founders Fund alone plans to invest $1 billion, the largest check ever written by the firm, two of the people said.
Anduril designs and builds autonomous systems and weapons for the military and other government agencies, including flying drones, missiles, underwater vessels and surveillance equipment for monitoring both national borders and the battlefield. It is one of a new wave of companies building systems based on A.I. technologies for the government.
Founders Fund, started by the entrepreneur and investor Peter Thiel, has backed Anduril since its start in 2017, and one of Anduril’s co-founders, Trae Stephens, is a partner at the firm. Mr. Thiel, who also co-founded Palantir, a military technology company, has long been a backer of Republican candidates, including President Trump in 2016 and JD Vance’s run for Senate in 2022.
Six months ago, Anduril raised $1.5 billion at a $14 billion valuation.
Founders Fund declined to comment. CNBC first reported details of the funding talks.
The latest influx of cash comes as defense technology start-ups are ebullient about their prospects. Enthusiasm for building technology for the U.S. military has grown in recent years in Silicon Valley, a reversal from more than a decade of shying away from those contracts. As recently as 2018, thousands of employees at Google signed a letter protesting the company’s military contracts.
That resistance has slowly shifted, as more venture capital firms pour money into the sector.
Mr. Trump is expected to turbocharge investments further. Palmer Luckey, Anduril’s founder, has supported the president since his 2016 campaign. He donated to Mr. Trump’s campaigns in the 2016, 2020 and 2024 elections, and has hosted fund-raisers.
On the night of the presidential election in November, Mr. Luckey posted a meme celebrating Mr. Trump’s victory. Elon Musk, the tech executive and a close adviser to the president, responded, saying it was “very important to open DoD/Intel to entrepreneurial companies like yours.”
Interview with Dr. Chaffra Affouda, Director of Business Development at Anduril Imaging - Photonics Project
Dr. Chaffra Affouda is a Business Development Director at Anduril Imaging, where his role is to grow the core technologies that will enable the next generation of infrared focal plane arrays and systems. He graduated from SUNY-Albany, Albany, NY in 2006 with a Ph.D. in Materials Science. He then...
photonicsproject.org
EXTRACT FROM THE INTERVIEW
Chaffra Affouda - Anduril Industries | LinkedIn
- Broad experience in product development & management of day and night vision… · Experience: Anduril Industries · Education: University at Albany, SUNY · Location: United States · 484 connections on LinkedIn. View Chaffra Affouda’s profile on LinkedIn, a professional community of 1 billion members.
Last edited:
Don't shoot the messenger. I'm an empty vessel making a lot of noise.
I think you're confusing Akida 2.0 with Akida 2.1.
It'll be very interesting to see if we partner up with Anduril in the not too distant future.
The reason why I say this is because I just did a little bit of poking around and discovered that Dr. Chaffra Affouda (see below), who is currently a Business Development Director at Anduril, worked as Senior Principal Systems Engineer at Raytheon Intelligence & Space from Apr 2019 to Sep 2022 before joinnig Anduril.
While he was at Raytheon, Dr. Affouda worked on developing multi-functional, neuromorphic infrared imaging sensors and cameras!
Prior to that he was with the US Naval Research Laboratory.
A.I. Military Start-Up Anduril Close to Deal That Would Value It at $28 Billion
The Southern California start-up, which builds flying drones and missiles, is set to raise up to $2.5 billion.
An employee demonstrates some of Anduril’s A.I. technology, in 2021.Credit...Philip Cheung for The New York Times
By Erin Griffith and Cade Metz
Reporting from San Francisco
Feb. 7, 2025
Sign up for the On Tech newsletter. Get our best tech reporting from the week. Get it sent to your inbox.
Anduril, an artificial intelligence military start-up, is set to complete a new round of funding that would double the value of the company to $28 billion, according to four people familiar with the negotiations.
The funding round, which is led by Founders Fund and has not yet closed, is raising up to $2.5 billion, the people said. Founders Fund alone plans to invest $1 billion, the largest check ever written by the firm, two of the people said.
Anduril designs and builds autonomous systems and weapons for the military and other government agencies, including flying drones, missiles, underwater vessels and surveillance equipment for monitoring both national borders and the battlefield. It is one of a new wave of companies building systems based on A.I. technologies for the government.
Founders Fund, started by the entrepreneur and investor Peter Thiel, has backed Anduril since its start in 2017, and one of Anduril’s co-founders, Trae Stephens, is a partner at the firm. Mr. Thiel, who also co-founded Palantir, a military technology company, has long been a backer of Republican candidates, including President Trump in 2016 and JD Vance’s run for Senate in 2022.
Six months ago, Anduril raised $1.5 billion at a $14 billion valuation.
Founders Fund declined to comment. CNBC first reported details of the funding talks.
The latest influx of cash comes as defense technology start-ups are ebullient about their prospects. Enthusiasm for building technology for the U.S. military has grown in recent years in Silicon Valley, a reversal from more than a decade of shying away from those contracts. As recently as 2018, thousands of employees at Google signed a letter protesting the company’s military contracts.
That resistance has slowly shifted, as more venture capital firms pour money into the sector.
Mr. Trump is expected to turbocharge investments further. Palmer Luckey, Anduril’s founder, has supported the president since his 2016 campaign. He donated to Mr. Trump’s campaigns in the 2016, 2020 and 2024 elections, and has hosted fund-raisers.
On the night of the presidential election in November, Mr. Luckey posted a meme celebrating Mr. Trump’s victory. Elon Musk, the tech executive and a close adviser to the president, responded, saying it was “very important to open DoD/Intel to entrepreneurial companies like yours.”
Interview with Dr. Chaffra Affouda, Director of Business Development at Anduril Imaging - Photonics Project
Dr. Chaffra Affouda is a Business Development Director at Anduril Imaging, where his role is to grow the core technologies that will enable the next generation of infrared focal plane arrays and systems. He graduated from SUNY-Albany, Albany, NY in 2006 with a Ph.D. in Materials Science. He then...
EXTRACT FROM THE INTERVIEW
View attachment 77435
Chaffra Affouda - Anduril Industries | LinkedIn
- Broad experience in product development & management of day and night vision… · Experience: Anduril Industries · Education: University at Albany, SUNY · Location: United States · 484 connections on LinkedIn. View Chaffra Affouda’s profile on LinkedIn, a professional community of 1 billion members.
View attachment 77436
That's the "Before" picture ...
This is "After":
"No ... sorry! That should have been to YOUR left."
Last edited:
Diogenese ............What a pity and a great loss, that AI_Inquirer has left this forum with his rather extensive knowledge and willingness to share information here at TSE ..... But was forced to leave due to some questionable posters here in the past having a go at him for not towing the Co line of only posting positive things about Akida.
I note, that back on the 24 Oct 2023 .... AI enquirer questioned where Akida 2 was up to...see below post of his:
- " Oct 24, 2023
- Something that has been particularly frustrating for me is prior to the launch of 2.0 it was repeated that EAP customers had already been actively evaluating 2.0, and the official release of 2.0 marked its general availability.
When the release came out it said 2.0 was available for ‘early access’ only.
Now in this 4C: “focusing on the ongoing development and availability of the 2nd Generation Akida technology platform to lead customers”.
I'm skeptical that they have a fully developed second-generation product, and the suggestion that customers were delaying decision on IP deals for the 2.0 release seems quite misleading. "......
Accordingly, imo it seems to me, that A1_Inquirer was once again correct on his assumptions from way back in Oct 2023 that Akida 2 was still on the development phase.
This in an August 2023 article from the Edge Impulse website.
Title:
By Adam TaylorAug 8, 2023
edgeimpulse.com
Title:
Benchmarking Akida with Edge Impulse: A Validation of Model Performance on BrainChip’s Akida Platform
Artificial IntelligenceBy Adam TaylorAug 8, 2023
Benchmarking Akida with Edge Impulse: A Validation of Model Performance on BrainChip's Akida Platform
See how BrainChip's Akida and Edge Impulse perform when it comes to visual wake word detection, keyword spotting, anomaly detection, and CIFAR classification.
edgeimpulse.com
From MegaChips recent presentation material(Ended 31 Dec, 2024)
Presentation Materials | MegaChips Corporation
We aim at sustainable growth through profitability and growth opportunity.
www.megachips.co.jp
CHIPS
Regular
View attachment 77441
From MegaChips recent presentation material(Ended 31 Dec, 2024)
Presentation Materials | MegaChips Corporation
We aim at sustainable growth through profitability and growth opportunity.www.megachips.co.jp
I don't see anything exciting or did I miss something here? They have a BrainChip license and have not used it so far. Right?
Where could it be used here?
Fullmoonfever
Top 20
I guess you have to remember that even if V2 was released to lead customers that they will run their own testing on scenarios or use cases that we may not have covered and as such will make ongoing development requests, additions or bug fixes.
Suggest you can view some of the timeline on GitHub, link below, where from about Sept 23 there has been regular activity as I read it anyway, on V2 models, layers etc etc.
You'll also find work done on 1500, 1000 and even a reference to 500, which I wonder if that was rebranded to Pico
I expect some of those are internal and others maybe externally raised.
Releases · Brainchip-Inc/akida_examples
Brainchip Akida Neuromorphic System-on-Chip examples and documentation. - Brainchip-Inc/akida_examples
github.com
Last edited:
Fullmoonfever
Top 20
Could be an interesting project to just keep an eye on.
Recently awarded (end last year) to a team from Purdue with other collaborators including Argonne which falls under the DOE.
We know the DOE understand Akida via Quantum Ventura and CyberNeuro-RT so maybe we may get some testing look ins with this project. And yes, I realise Dr. Kristofor Carlson has links back to Purdue too as well as Global Foundry, our 22nm producer, also having a partnership with Purdue.
microelectronicscommons.org
Artificial Intelligence (AI) Hardware ($8.7M)
CHEETA: CMOS+MRAM Hardware for Energy-EfficienT AI
This project seeks to develop a neuromorphic processor with in-memory computing (IMC) to overcome the von Neumann bottleneck and MRAM for higher density and energy efficiency that enables a new generation of robust energy efficient AI. The desired end-state will see a greater than 100X improvement in energy efficiency and sensor-to-decision latency over current commercial state-of-the-art solutions.
www.purdue.edu
Recently awarded (end last year) to a team from Purdue with other collaborators including Argonne which falls under the DOE.
We know the DOE understand Akida via Quantum Ventura and CyberNeuro-RT so maybe we may get some testing look ins with this project. And yes, I realise Dr. Kristofor Carlson has links back to Purdue too as well as Global Foundry, our 22nm producer, also having a partnership with Purdue.
Silicon Crossroads Microelectronics Commons Hub - Microelectronics Commons
Silicon Crossroads Microelectronics Commons Hub (SCMC), led by the Applied Research Institute, is a dynamic innovation ecosystem composed of diverse partners committed to expanding America’s microelectronics footprint. By fostering strong collaboration, SCMC strategically supports innovation...
microelectronicscommons.org
Artificial Intelligence (AI) Hardware ($8.7M)
CHEETA: CMOS+MRAM Hardware for Energy-EfficienT AI
This project seeks to develop a neuromorphic processor with in-memory computing (IMC) to overcome the von Neumann bottleneck and MRAM for higher density and energy efficiency that enables a new generation of robust energy efficient AI. The desired end-state will see a greater than 100X improvement in energy efficiency and sensor-to-decision latency over current commercial state-of-the-art solutions.
Purdue University wins Microelectronics Commons Project to advance AI hardware through the Silicon Crossroads Microelectronics Commons Hub
A Purdue project through the Microelectronic Commons program will pursue lab-to-fab transition of pioneering research performed by the university Purdue for over a decade.
www.purdue.edu
Hi DB,Could be an interesting project to just keep an eye on.
Recently awarded (end last year) to a team from Purdue with other collaborators including Argonne which falls under the DOE.
We know the DOE understand Akida via Quantum Ventura and CyberNeuro-RT so maybe we may get some testing look ins with this project. And yes, I realise Dr. Kristofor Carlson has links back to Purdue too as well as Global Foundry, our 22nm producer, also having a partnership with Purdue.
Silicon Crossroads Microelectronics Commons Hub - Microelectronics Commons
Silicon Crossroads Microelectronics Commons Hub (SCMC), led by the Applied Research Institute, is a dynamic innovation ecosystem composed of diverse partners committed to expanding America’s microelectronics footprint. By fostering strong collaboration, SCMC strategically supports innovation...
Artificial Intelligence (AI) Hardware ($8.7M)
CHEETA: CMOS+MRAM Hardware for Energy-EfficienT AI
This project seeks to develop a neuromorphic processor with in-memory computing (IMC) to overcome the von Neumann bottleneck and MRAM for higher density and energy efficiency that enables a new generation of robust energy efficient AI. The desired end-state will see a greater than 100X improvement in energy efficiency and sensor-to-decision latency over current commercial state-of-the-art solutions.
Purdue University wins Microelectronics Commons Project to advance AI hardware through the Silicon Crossroads Microelectronics Commons Hub
A Purdue project through the Microelectronic Commons program will pursue lab-to-fab transition of pioneering research performed by the university Purdue for over a decade.
This is analog:
Purdue’s project — CHEETA: CMOS+MRAM Hardware for Energy-EfficienT AI — will adopt a CMOS+X approach, specifically leveraging the unique capabilities of magnetic random-access memory (MRAM) to design efficient in-memory computing hardware fabrics.
In theory analog would be more effecient and have lower latency than digital, but manufacturing variablity and temperature fluctuations tend to dispel these advantages. Of course, there is always the possibility that ongoing research can improve the performance of analog NNs.
On the other hand, there was a comment (TL?) in the recent EETimes podcast about not needing fancy/non-standard fab tech.
Fullmoonfever
Top 20
I'll defer to your experience in this D however I recalled Numem were developing MRAM with Akida so figured it may (or maybe not) be feasible in some way.
From their Ph I on Techport.
Numem proposes in Phase-I to create a interface system with MRAM which can connect with AKIDA Neuromorphic Processor from Brainchip to either limit or replace FLASH operations with MRAM.
NASA TechPort
NASA's Technology Portfolio Management System (TechPort) is a single, comprehensive resource for locating detailed information about NASA-funded technologies. Those technologies cover a broad range of areas, such as propulsion, nanotechnology, robotics, and human health. You can find useful...
Yes - that does ring a bell - well remembered.I'll defer to your experience in this D however I recalled Numem were developing MRAM with Akida so figured it may (or maybe not) be feasible in some way.
From their Ph I on Techport.
Numem proposes in Phase-I to create a interface system with MRAM which can connect with AKIDA Neuromorphic Processor from Brainchip to either limit or replace FLASH operations with MRAM.
NASA TechPort
NASA's Technology Portfolio Management System (TechPort) is a single, comprehensive resource for locating detailed information about NASA-funded technologies. Those technologies cover a broad range of areas, such as propulsion, nanotechnology, robotics, and human health. You can find useful...techport.nasa.gov
I recall when I read this the first time, my understanding was that the MRAM would be more of an off-processor chip store for the boot code and weight storage which, in my casual reading, I took to mean weight backup.
However, the reference to "continuous learning" does suggest a closer involvement.
Weighing against that is the reference to an MRAM interface which can connect to the Akida neuromorphic processor, and I gave more weight to this express statement, so I'm still inclined to the backup interpretation. It would require a total redesign of Akida and an entirely different manufacturing process to replace Akida's SRAM with MRAM, which would be well outside the budget of an SBIR.
Space applications require FLASH memory for boot and weight storage in case of power loss or intermittent power failures. The FLASH memory has limitations on speed and life-time is limited by about 1M cycles of memory operations due to its endurance. For Deep Space Missions where continuous learning is required with updates on the non-volatile memory, a robust radiation tolerant memory with SRAM like performance but still with non volatility and high endurance is required. MRAM which offers 2.5X to 3.5X density advantage over SRAM, 1000X better endurance over FLASH, high radiation tolerance above 100Krad to 1Mrad and ultra-low power standby leakage which is critical for long battery life between solar recharge is a big advantage for these critical SPACE missions. Numem proposes in Phase-I to create a interface system with MRAM which can connect with AKIDA Neuromorphic Processor from Brainchip to either limit or replace FLASH operations with MRAM.
As an aside, this also confirms something else I was pondering - that SNNs are more fault tolerant than arithmetic-based processors, my supposition being that this is because SNNs work on probability rather than mathematical precision, so the odd random fault is unlikely to change the probabilities much.
" Neuromorphic architectures are inherently fault tolerant,"
Last edited:
Maybe Nintendo could release a new game - Brainchip Investing. The game concept is that strange snippets of information are released and you have to join the dots in real time. Crazyiest dot joiner wins. Slowest dot joiner loses.
KiKi
Regular
Pom down under
Top 20
Expectations on getting some real $$$ news this week.
Last edited:
Chart analysts in the HC forum are like,
“When the coyote runs through the prairie, the rabbit hole is south of the northern spaghetti stalk. Meanwhile, the farmer had five beers at the tavern while his wife and the son’s daughter were watching the rain from the café during the summer sale. Star Trek doesn’t get any more interesting either. … Just my opinion, though—things could turn out completely different.”
That just shows they have no clue… as if the woman would simply let the man drink five beers. She would force him to come along—which, for me, would be a bearish sign.
“When the coyote runs through the prairie, the rabbit hole is south of the northern spaghetti stalk. Meanwhile, the farmer had five beers at the tavern while his wife and the son’s daughter were watching the rain from the café during the summer sale. Star Trek doesn’t get any more interesting either. … Just my opinion, though—things could turn out completely different.”
That just shows they have no clue… as if the woman would simply let the man drink five beers. She would force him to come along—which, for me, would be a bearish sign.
Similar threads
