BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
A bit of Edge AI news:
https://www.msn.com/en-au/money/mar...S&cvid=68c7e3bb0afb440caea0bead03f07e88&ei=47
...
Waymo, the self-driving taxi service based in the US, is expanding its fleet of autonomous taxis to 3500 vehicles by 2026.
Kodiak Robotics is pushing into the long-haul autonomous trucking sector, coaxed along by a US$50 million grant from the US Department of Defense.
For all that, edge AI computing is still in its infancy.
The technology is undergoing rapid, almost daily advancement, but the challenges are many-fold.
“Creating something that’s fit for purpose and delivers value to customers at an attractive price point is a real technical challenge,” Harvest Technology (ASX:HTG) chief product officer Damiain Brown told Stockhead.
Harvest Tech is an operations solutions technology company, specialising in secure and stable remote and off-shore applications.
The company’s main offering is the Nodestream Protocol, a remote monitoring and operations product that offers huge bandwidth efficiency and connection stability advantages over traditional operational solutions.
“Our customers range widely. Some companies want AI in everything but have no budget, while defence contracts might have very deep pockets but very specific standards to meet.
“It’s difficult to design an agnostic system that will be fit for both purposes.”
“We work in really remote offshore sites. They’re disconnected. They don’t have the networks and everything else that people who operate in metropolitan areas have,” Brown explained.
“The main NEON platform will work around intelligent monitoring of remote sites, with a set of actions and commands the system can enact on top of that.
“You can think of it like a digital watchdog linked with a concierge intelligent enough to respond to incidents.”
Harvest imagines deploying intelligent operational incident monitors capable of not only recognising emergency situations, but deploying alarms, response systems and alerting the necessary stakeholders in real time, even in the most remote locations.
The technology could respond to incidents on unmanned vessels, enable firefighters to communicate and coordinate deep within inaccessible mountain ranges, or even monitor pollution and fishing nets via millions of individual deep-sea sensors.
That’s the goal, but there’s still a lot to accomplish before the technology can be fully realised in these ways.
Meanwhile hundreds of thousands of individual companies are developing proprietary offerings in the edge AI space, all racing each other to be the next big thing.
“There are so many versions of this technology, [so when] choosing which one to integrate for a broad swathe of industries and applications – everything from hydrocarbon production to first responders or defence – being agnostic is really important,” Brown said.
“We’ve also got to future proof. A lot of what’s available today will be legacy within weeks or months.
“This technology is just moving so damn quick.”
Brainchip Holdings (ASX:BRN) is quite literally developing the chips that could power Harvest’s edge AI goals.
A self-described ‘worldwide leader in edge AI on-chip processing’, Brainchip is developing its Akida technology to mimic a human brain, analysing sensor inputs at the point of acquisition (the chip itself) and processing the data on the spot.
BRN’s core offering is the Akida AI Acceleration Processor, a low-power, real-time AI processing chip designed to facilitate vision, audio and sensor functions.
It’s use applications include things like vehicle health monitoring via engine noises, vibrations and brake wear, and radar and LiDAR-based low-visibility perception for autonomous vehicles and similar technology.
Still on the chip architecture front, Weebit Nano (ASX:WBT) is developing a new class of memory chips called ReRAM, which promise to overcome current chip limitations in terms of power consumption, speed, endurance, cost, and size.
Resistive random-access memory (ReRAM) was one of the centre pieces of the 2024 TSMC Technology Symposium, positioned as a Flash memory chip replacement and a leading contender for machine learning applications.
The technology is incredibly technically advanced, but it boils down to being able to squeeze a hell of a lot more data onto much smaller chips, with lightning-fast data access at a fraction of the cost.
Weebit isn’t the only one developing ReRAM, but it’s managed to create a bit of space for itself by designing its chips with standard materials and tools, meaning they can be manufactured in existing plants with very little investment or equipment changes required.
Moving away from chip architecture, DXN Solutions (ASX
XN) has a novel solution for edge computing needs – miniaturised, modular data centres (DCs).
Rather than concentrating processing power in massive, power-hungry, centralised DCs, DXN is building bespoke, prefabricated Edge Data Centres.
Specifically designed for Edge computing, DXN offers custom-sized, modular data centres of all sizes, ranging from a simple broom closet-sized two-rack micro data centre, to 50-rack, warehouse-sized DCs.
DXN has already inked several data centre delivery deals, including with major gold miner Newcrest Mining (ASX:NCM) and international construction contractor Multiplex.
The company has deployed its edge AI solution to more than 30 locations across Australia, the South Pacific and Africa, servicing companies like Boeing and Anglo America.
That’s just one, relatively small ASX-listed company – uptake of Edge AI has been incredibly fast, growing at an incredible rate in terms of market penetration.
SHD Group, a global technology and automotive-focused market analysis firm, predicts edge AI will penetrate more than 31% of the greater market by 2030, foreseeing revenues of US$100 billion by the same year and a massive 55% capture of the overall AI market.
That’s part of the challenge for companies such as Harvest Tech, which are deliberately reaching just a little further than our current capabilities.
“We’re trying to find the thin edge of the wedge and build for tomorrow,” Brown explained.
“We’ve got to be very careful not to paint ourselves into a corner. Obsolescence is a real concern.
“Once, you’d build something and it would be fit for purpose for years. Now, it could be disrupted and obsolete within weeks.”
In other words, tread carefully, and don’t trust the PR spin. Things are moving far too quickly in the edge AI space for anyone to get too comfortable.
That said, all signs are pointing to a far more connected, efficient and autonomous future just over the horizon.
“Edge AI is the ultimate and probably only scalable way to do AI in the real world – collecting, analysing, and acting on data where it lives,” Qualcomm senior director of AI Research Evgeni Gousev said in an interview with Schneider Electric.
“In our industry, with the way things progress now, one day equals a year of development,” Gousev said at the tinyML Innovation Forum in Milan.
“It’s like a snowball effect.”
TinyML Foundation executive director Pete Bernard said that while Edge AI isn’t as exciting or glamorous as GenAI, it’s having a real impact on people’s lives.
“The value proposition is pretty universal,” he said.
“Whether you’re using it to fix a water system or grow better crops or have better health outcomes, its lower cost, lower power requirements … people can get their hands on this stuff and just start building.
“Every Fortune 500 chief technology officer right now has an AI strategy, and I suspect edge AI is going to be part of that strategy.”
https://www.msn.com/en-au/money/mar...S&cvid=68c7e3bb0afb440caea0bead03f07e88&ei=47
Beyond the clouds: Bringing the power of edge AI to the far reaches of society
Story by Phoebe Shields...
Are we there yet?
Less than a week ago, a Tesla drove itself from Sydney to Melbourne with only passive supervision from a human driver.Waymo, the self-driving taxi service based in the US, is expanding its fleet of autonomous taxis to 3500 vehicles by 2026.
Kodiak Robotics is pushing into the long-haul autonomous trucking sector, coaxed along by a US$50 million grant from the US Department of Defense.
For all that, edge AI computing is still in its infancy.
The technology is undergoing rapid, almost daily advancement, but the challenges are many-fold.
“Creating something that’s fit for purpose and delivers value to customers at an attractive price point is a real technical challenge,” Harvest Technology (ASX:HTG) chief product officer Damiain Brown told Stockhead.
Harvest Tech is an operations solutions technology company, specialising in secure and stable remote and off-shore applications.
The company’s main offering is the Nodestream Protocol, a remote monitoring and operations product that offers huge bandwidth efficiency and connection stability advantages over traditional operational solutions.
“Our customers range widely. Some companies want AI in everything but have no budget, while defence contracts might have very deep pockets but very specific standards to meet.
“It’s difficult to design an agnostic system that will be fit for both purposes.”
Going fully robotic
Harvest Tech dreams of facilitating fully autonomous remote worksites via its Nodestream protocol, and the company is pushing strongly into edge AI within its Nodestream Enhanced Operating Network (NEON) development project to facilitate that vision.“We work in really remote offshore sites. They’re disconnected. They don’t have the networks and everything else that people who operate in metropolitan areas have,” Brown explained.
“The main NEON platform will work around intelligent monitoring of remote sites, with a set of actions and commands the system can enact on top of that.
“You can think of it like a digital watchdog linked with a concierge intelligent enough to respond to incidents.”
Harvest imagines deploying intelligent operational incident monitors capable of not only recognising emergency situations, but deploying alarms, response systems and alerting the necessary stakeholders in real time, even in the most remote locations.
The technology could respond to incidents on unmanned vessels, enable firefighters to communicate and coordinate deep within inaccessible mountain ranges, or even monitor pollution and fishing nets via millions of individual deep-sea sensors.
That’s the goal, but there’s still a lot to accomplish before the technology can be fully realised in these ways.
Meanwhile hundreds of thousands of individual companies are developing proprietary offerings in the edge AI space, all racing each other to be the next big thing.
“There are so many versions of this technology, [so when] choosing which one to integrate for a broad swathe of industries and applications – everything from hydrocarbon production to first responders or defence – being agnostic is really important,” Brown said.
“We’ve also got to future proof. A lot of what’s available today will be legacy within weeks or months.
“This technology is just moving so damn quick.”
Who’s doing the leg work?
While Harvest Tech works to bring its end-use vision of autonomous deep-sea rescue missions and fully remote oil rig repairs to life, a swathe of other ASX-listed companies are working on the foundational technology to facilitate it.Brainchip Holdings (ASX:BRN) is quite literally developing the chips that could power Harvest’s edge AI goals.
A self-described ‘worldwide leader in edge AI on-chip processing’, Brainchip is developing its Akida technology to mimic a human brain, analysing sensor inputs at the point of acquisition (the chip itself) and processing the data on the spot.
BRN’s core offering is the Akida AI Acceleration Processor, a low-power, real-time AI processing chip designed to facilitate vision, audio and sensor functions.
It’s use applications include things like vehicle health monitoring via engine noises, vibrations and brake wear, and radar and LiDAR-based low-visibility perception for autonomous vehicles and similar technology.
Still on the chip architecture front, Weebit Nano (ASX:WBT) is developing a new class of memory chips called ReRAM, which promise to overcome current chip limitations in terms of power consumption, speed, endurance, cost, and size.
Resistive random-access memory (ReRAM) was one of the centre pieces of the 2024 TSMC Technology Symposium, positioned as a Flash memory chip replacement and a leading contender for machine learning applications.
The technology is incredibly technically advanced, but it boils down to being able to squeeze a hell of a lot more data onto much smaller chips, with lightning-fast data access at a fraction of the cost.
Weebit isn’t the only one developing ReRAM, but it’s managed to create a bit of space for itself by designing its chips with standard materials and tools, meaning they can be manufactured in existing plants with very little investment or equipment changes required.
Moving away from chip architecture, DXN Solutions (ASX
XN) has a novel solution for edge computing needs – miniaturised, modular data centres (DCs).Rather than concentrating processing power in massive, power-hungry, centralised DCs, DXN is building bespoke, prefabricated Edge Data Centres.
Specifically designed for Edge computing, DXN offers custom-sized, modular data centres of all sizes, ranging from a simple broom closet-sized two-rack micro data centre, to 50-rack, warehouse-sized DCs.
DXN has already inked several data centre delivery deals, including with major gold miner Newcrest Mining (ASX:NCM) and international construction contractor Multiplex.
The company has deployed its edge AI solution to more than 30 locations across Australia, the South Pacific and Africa, servicing companies like Boeing and Anglo America.
That’s just one, relatively small ASX-listed company – uptake of Edge AI has been incredibly fast, growing at an incredible rate in terms of market penetration.
SHD Group, a global technology and automotive-focused market analysis firm, predicts edge AI will penetrate more than 31% of the greater market by 2030, foreseeing revenues of US$100 billion by the same year and a massive 55% capture of the overall AI market.
On the edge of tomorrow
So, the technology isn’t quite here yet, but at current rates of progress it promises to arrive at a blistering pace.That’s part of the challenge for companies such as Harvest Tech, which are deliberately reaching just a little further than our current capabilities.
“We’re trying to find the thin edge of the wedge and build for tomorrow,” Brown explained.
“We’ve got to be very careful not to paint ourselves into a corner. Obsolescence is a real concern.
“Once, you’d build something and it would be fit for purpose for years. Now, it could be disrupted and obsolete within weeks.”
In other words, tread carefully, and don’t trust the PR spin. Things are moving far too quickly in the edge AI space for anyone to get too comfortable.
That said, all signs are pointing to a far more connected, efficient and autonomous future just over the horizon.
“Edge AI is the ultimate and probably only scalable way to do AI in the real world – collecting, analysing, and acting on data where it lives,” Qualcomm senior director of AI Research Evgeni Gousev said in an interview with Schneider Electric.
“In our industry, with the way things progress now, one day equals a year of development,” Gousev said at the tinyML Innovation Forum in Milan.
“It’s like a snowball effect.”
TinyML Foundation executive director Pete Bernard said that while Edge AI isn’t as exciting or glamorous as GenAI, it’s having a real impact on people’s lives.
“The value proposition is pretty universal,” he said.
“Whether you’re using it to fix a water system or grow better crops or have better health outcomes, its lower cost, lower power requirements … people can get their hands on this stuff and just start building.
“Every Fortune 500 chief technology officer right now has an AI strategy, and I suspect edge AI is going to be part of that strategy.”
Stable Genius
Regular
It’d be nice to be partnered up with Mirle from Taiwan who’s building robots and autonomous quadraped robots.
brainchip.com
BrainChip Collaborates with Chelpis-Mirle on Security Solution
BrainChip teams with Chelpis and Mirle to deliver next-generation security solutions powered by low-power, adaptive edge AI capabilities.
brainchip.com
Last edited:
Getupthere
Regular
Beyond the clouds: Bringing the power of edge AI to the far reaches of society — Stockhead
Edge AI promises real-time data processing on local devices, powering faster decisions and smarter automation. It’s set to drive over half the AI market by 2030.
“Every Fortune 500 chief technology officer right now has an AI strategy, and I suspect edge AI is going to be part of that strategy.”
Last edited:
#neuromorphiccomputing #braininspired #ai #lowpowercomputing #parallelprocessing #spikingneuralnetworks #edgeai #innovation #futuretech #computingevolution | Chandra Sekhar Pinninti
🧠 Neuromorphic Computing: Computers That Think Like Brains Q: What's wrong with current computers? A: They're energy hogs! Your brain uses 20 watts (like a light bulb) but current AI needs thousands of watts. Plus they process everything step-by-step instead of parallel thinking. Q: What is...
Pom down under
Top 20
So who paid to produce that Trim capital research report?
Pom down under
Top 20
Food4 thought
Regular
Sean did out of his own pocket so to keep the price low so he can get more shares at a low price for his bonus lol
No idea
Not advice or correct information
Pom down under
Top 20
I think there could be a more sinister reason behind it.Maybe rob telson! It’s personal
BrainChip's Ritik Shrivastava presents at Interspeech 2025 | BrainChip posted on the topic | LinkedIn
Last month, our colleague Ritik Shrivastava, Solutions Architect at BrainChip, attended Interspeech 2025 in Rotterdam. Interspeech is the world’s largest conference for spoken language processing. This year’s theme, “Fair and Inclusive Speech Science and Technology,” emphasized how diversity in...
aTENNuate: Real-Time Audio Denoising
aTENNuate: Real-Time Audio Denoising by Ritik Shrivastava, Solutions Architect BrainChip's aTENNuate is a new audio denoising solution that uses a type of deep learning
brainchip.com
Pom down under
Top 20
The penny' dropped ...
Tony reportedly said that originally TENNs couldn't do recurrence.
So that meant that TENNs needed to be tied to the original Akida 1 NPU. - "Better with Akida".
But Tony also intimated that TENNs was sufficiently evolved to do recurrence by the time GenAI/Akida 3 were developed. He has also indicated that there is some (minimal) CPU involvement in the operation of TENNs, as distinct from AKida NPUs which do not require processor involvement in detection/inference/classification.
It was only after the announcement of GenAI/Akida 3 that we saw Akida 1 being spoken of only in terms of MACs, with no mention of the old NPUs.
So, when TENNs learned recurrence, was this the tipping point which meant that Akida 1 had been superseded?
I wonder what hardware improvements were involved in implementing recurrence with TENNs. I imagine that memory would be involved. In the Roadmap, JT mentioned that patents for improvements in memory operation were in the pipeline.
This is the patent application implementing recurrence:
US2025209313A1 METHOD AND SYSTEM FOR IMPLEMENTING ENCODER PROJECTION IN NEURAL NETWORKS 20231222 pub 250625
[0027] FIG. 10 is a flow chart of a method performed by the neural processor for performing, in the recurrent mode, encoder projection, mid-layer processing, and decoding which generates content using one or more neural network layers of the neural network in accordance with some embodiments.
The patent was filed in December 2023. Edit: The original TENNs patent was filed in mid-2022:
WO2023250093A1 METHOD AND SYSTEM FOR IMPLEMENTING TEMPORAL CONVOLUTION IN SPATIOTEMPORAL NEURAL NETWORKS 20220622
... so the recurrent feature was the result of 18 months further development.
The patent application would have preceded the final circuit design. It may have been breadboarded in FPGA before th4e ASIC design was done for the commercial IP release.
Tony reportedly said that originally TENNs couldn't do recurrence.
So that meant that TENNs needed to be tied to the original Akida 1 NPU. - "Better with Akida".
But Tony also intimated that TENNs was sufficiently evolved to do recurrence by the time GenAI/Akida 3 were developed. He has also indicated that there is some (minimal) CPU involvement in the operation of TENNs, as distinct from AKida NPUs which do not require processor involvement in detection/inference/classification.
It was only after the announcement of GenAI/Akida 3 that we saw Akida 1 being spoken of only in terms of MACs, with no mention of the old NPUs.
So, when TENNs learned recurrence, was this the tipping point which meant that Akida 1 had been superseded?
I wonder what hardware improvements were involved in implementing recurrence with TENNs. I imagine that memory would be involved. In the Roadmap, JT mentioned that patents for improvements in memory operation were in the pipeline.
This is the patent application implementing recurrence:
US2025209313A1 METHOD AND SYSTEM FOR IMPLEMENTING ENCODER PROJECTION IN NEURAL NETWORKS 20231222 pub 250625
[0027] FIG. 10 is a flow chart of a method performed by the neural processor for performing, in the recurrent mode, encoder projection, mid-layer processing, and decoding which generates content using one or more neural network layers of the neural network in accordance with some embodiments.
The patent was filed in December 2023. Edit: The original TENNs patent was filed in mid-2022:
WO2023250093A1 METHOD AND SYSTEM FOR IMPLEMENTING TEMPORAL CONVOLUTION IN SPATIOTEMPORAL NEURAL NETWORKS 20220622
... so the recurrent feature was the result of 18 months further development.
The patent application would have preceded the final circuit design. It may have been breadboarded in FPGA before th4e ASIC design was done for the commercial IP release.
Last edited:
Similar threads
