BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
Fullmoonfever
Top 20
Hadn't personally seen this project over at Edge Impulse before.
Google search link said late May 23 but who knows.
Have taken all the code sections etc out but full read at the link. Pretty cool running with FOMO.
A computer vision project that monitors vehicle traffic in real-time using video inferencing performed on the Brainchip Akida Development Kit.
Created By: Naveen Kumar
Public Project Link: https://studio.edgeimpulse.com/public/222419/latest
By capturing moving vehicle images, aerial cameras can provide information about traffic conditions, which is beneficial for governments and planners to manage traffic and enhance urban mobility. Detecting moving vehicles with low-powered devices is still a challenging task. We are going to tackle this problem using a Brainchip Akida neural network accelerator.
The kit consists of an Akida PCie board, a Raspberry Pi Compute Module 4 with Wi-Fi and 8 GB RAM, and a Raspberry Pi Compute Module 4 I/O Board. The disassembled kit is shown below.
The Akida PCIe board can be connected to the Raspberry Pi Compute Module 4 IO Board through the PCIe Gen 2 x1 socket available onboard.
The FOMO model uses an architecture similar to a standard image classification model which splits the input image into a grid and runs the equivalent of image classification across all cells in the grid independently in parallel. By default the grid size is 8x8 pixels, which means for a 224x224 image, the output will be 28x28 as shown in the image below.
For localization, it cuts off the last layers of the classification model and replaces this layer with a per-region class probability map, and subsequently applies a custom loss function that forces the network to fully preserve the locality in the final layer. This essentially gives us a heat map of vehicle locations. FOMO works on the constraining assumption that all of the bounding boxes are square, have a fixed size, and the objects are spread over the output grid. In the aerial view images, vehicles look similar in size hence FOMO works quite well.
Google search link said late May 23 but who knows.
Have taken all the code sections etc out but full read at the link. Pretty cool running with FOMO.
Traffic Monitoring using the Brainchip Akida Neuromorphic Processor
A computer vision project that monitors vehicle traffic in real-time using video inferencing performed on the Brainchip Akida Development Kit.
Created By: Naveen Kumar
Public Project Link: https://studio.edgeimpulse.com/public/222419/latest
Overview
A highly efficient computer-vision system that can detect moving vehicles with great accuracy and relative motion, all while consuming minimal power.
By capturing moving vehicle images, aerial cameras can provide information about traffic conditions, which is beneficial for governments and planners to manage traffic and enhance urban mobility. Detecting moving vehicles with low-powered devices is still a challenging task. We are going to tackle this problem using a Brainchip Akida neural network accelerator.
Hardware Selection
In this project, we'll utilize BrainChip’s Akida Development Kit. BrainChip's neuromorphic processor IP uses event-based technology for increased energy efficiency. It allows incremental learning and high-speed inference for various applications, including convolutional neural networks, with exceptional performance and low power consumption.The kit consists of an Akida PCie board, a Raspberry Pi Compute Module 4 with Wi-Fi and 8 GB RAM, and a Raspberry Pi Compute Module 4 I/O Board. The disassembled kit is shown below.
The Akida PCIe board can be connected to the Raspberry Pi Compute Module 4 IO Board through the PCIe Gen 2 x1 socket available onboard.
The FOMO model uses an architecture similar to a standard image classification model which splits the input image into a grid and runs the equivalent of image classification across all cells in the grid independently in parallel. By default the grid size is 8x8 pixels, which means for a 224x224 image, the output will be 28x28 as shown in the image below.
For localization, it cuts off the last layers of the classification model and replaces this layer with a per-region class probability map, and subsequently applies a custom loss function that forces the network to fully preserve the locality in the final layer. This essentially gives us a heat map of vehicle locations. FOMO works on the constraining assumption that all of the bounding boxes are square, have a fixed size, and the objects are spread over the output grid. In the aerial view images, vehicles look similar in size hence FOMO works quite well.
Confusion Matrix
Once the training is completed we can see the confusion matrices as shown below. By using the post-training quantization, the Convolutional Neural Networks (CNN) are converted to a low-latency and low-power Spiking Neural Network (SNN) for use with the Akida runtime. We can see in the below image, the F1 score of 94% of the Quantized (Akida) model is better than that of the Quantized (int8) model.Demo
The video used for the demonstration runs at a framerate of 24 fps, and the inferencing takes approximately 40ms per frame, resulting in real-time inferencing.Conclusion
This project highlights the impressive abilities of the Akida PCIe board. Boasting low power consumption, it could be used as a highly effective device for real-time object detection in various industries for numerous use cases.
rgupta
Regular
Intel Unveils Aurora genAI: A Trillion-Parameter AI Model to Revolutionize Scientific Breakthroughs and Predict the Unseen
At the ISC23 keynote, Intel announced Aurora genAI – a science-focused generative AI model with a trillion parameters, almost six times more than in the free and public versions of ChatGPT. This news has sparked conversations about all the possibilities this model can unlock. It has always been...
www.marktechpost.com
rgupta
Regular
Eric Feuilleaubois (Ph.D) on LinkedIn: Speed up neural networks 3.3x with Sparse Weight Activation Training…
Speed up neural networks 3.3x with Sparse Weight Activation Training [Breakdowns] https://buff.ly/3qHx31N
MadMayHam | 合氣道
Regular
Ahhh... so that's what all the silicon is for!
CrabmansFriend
Regular
Nothing specific about Brainchip, nonetheless an interesting article in my opinion:
Favorite Quotes:
Ian Cutress:
“The number that I always get quoted is something like 90% of the training market is currently hosted by Nvidia. But when I speak training, there’s obviously the whole world of inference that sometimes we forget about.”
...
However, there are many more real-world needs for inference, and Nvidia has no plans to meet them, he said.
...
“The devices that we hold in our hands, the devices on the edge and even going in to solve the data center market, there’s a lot more malleability there for these new AI hardware vendors to play in, to take advantage of, to find cost-effective solutions—and optimize solutions with customers,” he said. “That’s where I see the biggest opportunity to kind of battle the Nvidia juggernaut.”
Bill Jenkins:
“I’ll go back to one of the biggest problems,” he said. “Not many people really know what they want to do. There are just so many ways and so many things that they could implement. You know, I look at the GPU, the CPU and even the FPGA as that flexible architecture that can handle everything. And then the question is: Does it need to do something really well, and is there an alternative dedicated piece of hardware for that something?”
Nitin Dahad:
Dahad pointed out that there is a lot of expertise required on the customer side in using the hardware and software for AI. He asked the panelists what they are asking for from the industry.
“I would say the No. 1 thing is, ‘I’ve got a model. How do I implement that on your architecture?’” Jenkins said. “And then they’re going to compare that performance against where they are today. So if somebody can provide [a product that is] going to be lower-latency, lower-power, higher-performance and turnkey … they’ll take it and tweak it over time.”
Favorite Quotes:
Ian Cutress:
“The number that I always get quoted is something like 90% of the training market is currently hosted by Nvidia. But when I speak training, there’s obviously the whole world of inference that sometimes we forget about.”
...
However, there are many more real-world needs for inference, and Nvidia has no plans to meet them, he said.
...
“The devices that we hold in our hands, the devices on the edge and even going in to solve the data center market, there’s a lot more malleability there for these new AI hardware vendors to play in, to take advantage of, to find cost-effective solutions—and optimize solutions with customers,” he said. “That’s where I see the biggest opportunity to kind of battle the Nvidia juggernaut.”
Bill Jenkins:
“I’ll go back to one of the biggest problems,” he said. “Not many people really know what they want to do. There are just so many ways and so many things that they could implement. You know, I look at the GPU, the CPU and even the FPGA as that flexible architecture that can handle everything. And then the question is: Does it need to do something really well, and is there an alternative dedicated piece of hardware for that something?”
Nitin Dahad:
Dahad pointed out that there is a lot of expertise required on the customer side in using the hardware and software for AI. He asked the panelists what they are asking for from the industry.
“I would say the No. 1 thing is, ‘I’ve got a model. How do I implement that on your architecture?’” Jenkins said. “And then they’re going to compare that performance against where they are today. So if somebody can provide [a product that is] going to be lower-latency, lower-power, higher-performance and turnkey … they’ll take it and tweak it over time.”
Last edited:
HopalongPetrovski
I'm Spartacus!
BrainChip and Socionext Provide a New Low-Power Artificial Intelligence Platform for AI Edge Applications.... Just not together, in the one package
hence Socionext also offers...
Look, I've tried to be positive with the following:
Astra Edge Camera
teksun.com
Gesture Control
teksun.com
40c was my sell line, but WTF...
Terroni2105
Founding Member
Hadn't personally seen this project over at Edge Impulse before.
Google search link said late May 23 but who knows.
Have taken all the code sections etc out but full read at the link. Pretty cool running with FOMO.
Traffic Monitoring using the Brainchip Akida Neuromorphic Processor
A computer vision project that monitors vehicle traffic in real-time using video inferencing performed on the Brainchip Akida Development Kit.
Created By: Naveen Kumar
Public Project Link: https://studio.edgeimpulse.com/public/222419/latest
Overview
A highly efficient computer-vision system that can detect moving vehicles with great accuracy and relative motion, all while consuming minimal power.
By capturing moving vehicle images, aerial cameras can provide information about traffic conditions, which is beneficial for governments and planners to manage traffic and enhance urban mobility. Detecting moving vehicles with low-powered devices is still a challenging task. We are going to tackle this problem using a Brainchip Akida neural network accelerator.
Hardware Selection
In this project, we'll utilize BrainChip’s Akida Development Kit. BrainChip's neuromorphic processor IP uses event-based technology for increased energy efficiency. It allows incremental learning and high-speed inference for various applications, including convolutional neural networks, with exceptional performance and low power consumption.
The kit consists of an Akida PCie board, a Raspberry Pi Compute Module 4 with Wi-Fi and 8 GB RAM, and a Raspberry Pi Compute Module 4 I/O Board. The disassembled kit is shown below.
View attachment 38629
The Akida PCIe board can be connected to the Raspberry Pi Compute Module 4 IO Board through the PCIe Gen 2 x1 socket available onboard.
View attachment 38628
The FOMO model uses an architecture similar to a standard image classification model which splits the input image into a grid and runs the equivalent of image classification across all cells in the grid independently in parallel. By default the grid size is 8x8 pixels, which means for a 224x224 image, the output will be 28x28 as shown in the image below.
View attachment 38627
For localization, it cuts off the last layers of the classification model and replaces this layer with a per-region class probability map, and subsequently applies a custom loss function that forces the network to fully preserve the locality in the final layer. This essentially gives us a heat map of vehicle locations. FOMO works on the constraining assumption that all of the bounding boxes are square, have a fixed size, and the objects are spread over the output grid. In the aerial view images, vehicles look similar in size hence FOMO works quite well.
Confusion Matrix
Once the training is completed we can see the confusion matrices as shown below. By using the post-training quantization, the Convolutional Neural Networks (CNN) are converted to a low-latency and low-power Spiking Neural Network (SNN) for use with the Akida runtime. We can see in the below image, the F1 score of 94% of the Quantized (Akida) model is better than that of the Quantized (int8) model.
Demo
The video used for the demonstration runs at a framerate of 24 fps, and the inferencing takes approximately 40ms per frame, resulting in real-time inferencing.
Conclusion
This project highlights the impressive abilities of the Akida PCIe board. Boasting low power consumption, it could be used as a highly effective device for real-time object detection in various industries for numerous use cases.
that’s great, I haven’t seen it before either, thanks for posting it
I think that encapsulates the quandary - back in the last millennium, marketing used to talk about customers' wants and customers' needs.
Because the capabilities of Akida are far beyond those of known technology, customers do not know what they want it to do, let alone what they need it to do.
CrabmansFriend
Regular
Although I would assume that the customers that don't know what they are looking for (yet), probably are in some part of the market that isn't that close to the edge. Every company that is designing/building sensors or uses certain sensors for specific aspects of their (hardware) product probably is aware of their top most areas to improve their product, e.g. their power budgets, latency, etc.
Companies in different sectors might just be trying to get a foot into this AI/ML thing and probably still need to figure out what they really want/need. At this state they would probably try to avoid mistakes and not to over optimize too much already in the beginning. I assume they would be looking for something more "general purpose" kind of, "you can run all the foundational models and then some" even if the performance could be better.
But maybe Renesas, ARM and the like of could help here until Brainchip becomes a brand in that space, gets recoginition for and/or verification by available end user/BTB products.
Fullmoonfever
Top 20
Interesting short article from the recent Autosens as an insight into the vehicle sensors required.
Also a snippet of a MB presso general statement. That's a shitload of data moving and how much latency is added?
Solutions like NVIDIA GPU AI may be one of the interim / transitional ones (imo) given their overall size, resources and supply ability however, as per a recent prev post of a conference paper, this processing comes at a substantial power cost.
There will come a point where the power consideration becomes a selling point of differentiation and / or an absolute necessity to end users / developers.
That's where we must be ready, developed and at the cutting edge after all this DD, testing etc that's been going on with EAP, continuing partnerships and potential early adopters.
semiengineering.com
JUNE 1ST, 2023 - BY: FRANK SCHIRRMEISTER
The recent AutoSens conference in Detroit left me questioning whether I should turn in my driver’s license. The answer the attending OEMs gave to all the discussions about the advantages of RGB cameras, ultrasound, radar, lidar, and thermal sensors was a unanimous “We probably need all of them in some form of combination” to make autonomy a reality in automotive. Together, these sensors are much better than my eyes and ears.
Technology progress is speedy in Automated Driving Assistance Systems (ADAS) and autonomous driving, but we are not there yet holistically. So I am keeping my license for some more time.
I traveled to Auto City to speak on a panel organized by Ann Mutschler that focused on the design chain aspects, together with Siemens EDA and GlobalFoundries. Ann’s write-up “Automotive Relationships Shifting With Chiplets” summarizes the panel well. The conference was a great experience as the networking allowed talking to the whole design chain from OEMs through Tier 1 system suppliers, Tier 2 semis and software developers, to Tier 3s like us in semiconductor IP. Given that the panel had a foundry, an IP vendor, and an EDA vendor, we quickly focused our discussions on chiplets.
Concerning the sensing aspects, Owl.Ai’s CEO and co-founder, Chuck Gershman, gave an excellent presentation summarizing the problem the industry is trying to solve – 700K annual worldwide pedestrian fatalities, 59% increase in pedestrian deaths in the last decade in the US and 76% of fatalities occurring at night. Government regulations are coming for pedestrian nighttime safety worldwide. Owl.Ai and Flir showcased thermal camera-related technologies, motivated by only 1 out of 23 vehicles passing all tests in a nighttime IIHS test using cameras and radar and on RGB image sensors not being able to see in complete darkness (just like me, I should say, but I am still keeping my driver’s license).
Source: Owl.Ai, AutoSens 2023, Detroit
Chuck nicely introduced the four critical phases of “detection” – is something there – “classification” – is it a person, car, or deer – “range estimation” – what distance in meters is the object – and “acting” – warning the driver or acting automatically. I liked Owl.Ai’s slide above, which shows the various sensing methods’ different use cases and flaws.
And during the discussion I had during the conference, the OEMs agreed that multiple sensors are needed.
Regarding the transition of driving from L3 to L4 robot taxis, Rivian’s Abdullah Zaidi showed the slide below outlining the different needs for cameras, radars, and lidars, and also the compute requirements.
Source: Rivian, AutoSens 2023, Detroit
No wonder automotive is such an attractive space for semiconductors. Computing, sensing, and data transport requirements are just growing enormously. And mind you that the picture above does not mention other cameras for in-cabin monitoring.
Besides the computing requirements, data transport is core to my day-to-day work. In one of his slides, Mercedes-Benz AG’s Konstantin Fichtner presented that the DrivePilot system records 33.73 GB of trigger measurements per minute – 281 times as much as it takes to watch a Netflix 4K stream. That’s a lot of data to transport across networks-on-chips (NoCs), between chips and chiplets. And it, of course, raises the question of on-board vs. off-board processing.
Are we there yet? Not quite, but we are getting closer. On the last day of the conference, Carnegie Mellon University’s Prof. Philip Koopman sobered up the audience with his talk “Defining Safety For Shared Human/Computer Driver Responsibility.” His keynote walked the audience through the accountability dilemma when a highly automated vehicle crashes and made some constructive suggestions on updating state laws. In their recently published essay “Winning the Imitation Game: Setting Safety Expectations for Automated Vehicles.” Prof. Koopman and William H. Widen from the University of Miami School of Law suggest that legislatures amend existing laws to create a new legal category of “computer driver” to allow a plaintiff to make a negligence claim.
To make that exact point, the universe created the situation the next day, which I took a picture of below. Can you see what’s wrong here?
Source: Frank Schirrmeister, May 2023
Yep, a pedestrian ghost in the machine.
To technology’s excuse, there had been a jay-walking pedestrian about 30 seconds ago, which probably erred on the side of caution. But still, this was a good reminder that future sensors are hopefully better than my eyes, and a thermal sensor would have helped here too.
All soberness and glitches aside, let’s not forget the end goal: Reducing fatalities due to traffic situations. And as I joked in my last blog on ISO 26262 safety, How Safe is Safe Enough: “If aliens would arrive and assess how to reduce traffic-related deaths, they would most certainly take humans off the streets.”
Brave new autonomous world, here we come. And I am keeping my license. That 1997 Miata doesn’t drive itself!
Also a snippet of a MB presso general statement. That's a shitload of data moving and how much latency is added?
Solutions like NVIDIA GPU AI may be one of the interim / transitional ones (imo) given their overall size, resources and supply ability however, as per a recent prev post of a conference paper, this processing comes at a substantial power cost.
There will come a point where the power consideration becomes a selling point of differentiation and / or an absolute necessity to end users / developers.
That's where we must be ready, developed and at the cutting edge after all this DD, testing etc that's been going on with EAP, continuing partnerships and potential early adopters.
How Many Senses Do You Need To Drive A Car?
Automotive computing, sensing, and data transport requirements are growing enormously.
semiengineering.com
How Many Senses Do You Need To Drive A Car?
Automotive computing, sensing, and data transport requirements are growing enormously.JUNE 1ST, 2023 - BY: FRANK SCHIRRMEISTER
The recent AutoSens conference in Detroit left me questioning whether I should turn in my driver’s license. The answer the attending OEMs gave to all the discussions about the advantages of RGB cameras, ultrasound, radar, lidar, and thermal sensors was a unanimous “We probably need all of them in some form of combination” to make autonomy a reality in automotive. Together, these sensors are much better than my eyes and ears.
Technology progress is speedy in Automated Driving Assistance Systems (ADAS) and autonomous driving, but we are not there yet holistically. So I am keeping my license for some more time.
I traveled to Auto City to speak on a panel organized by Ann Mutschler that focused on the design chain aspects, together with Siemens EDA and GlobalFoundries. Ann’s write-up “Automotive Relationships Shifting With Chiplets” summarizes the panel well. The conference was a great experience as the networking allowed talking to the whole design chain from OEMs through Tier 1 system suppliers, Tier 2 semis and software developers, to Tier 3s like us in semiconductor IP. Given that the panel had a foundry, an IP vendor, and an EDA vendor, we quickly focused our discussions on chiplets.
Concerning the sensing aspects, Owl.Ai’s CEO and co-founder, Chuck Gershman, gave an excellent presentation summarizing the problem the industry is trying to solve – 700K annual worldwide pedestrian fatalities, 59% increase in pedestrian deaths in the last decade in the US and 76% of fatalities occurring at night. Government regulations are coming for pedestrian nighttime safety worldwide. Owl.Ai and Flir showcased thermal camera-related technologies, motivated by only 1 out of 23 vehicles passing all tests in a nighttime IIHS test using cameras and radar and on RGB image sensors not being able to see in complete darkness (just like me, I should say, but I am still keeping my driver’s license).
Source: Owl.Ai, AutoSens 2023, Detroit
Chuck nicely introduced the four critical phases of “detection” – is something there – “classification” – is it a person, car, or deer – “range estimation” – what distance in meters is the object – and “acting” – warning the driver or acting automatically. I liked Owl.Ai’s slide above, which shows the various sensing methods’ different use cases and flaws.
And during the discussion I had during the conference, the OEMs agreed that multiple sensors are needed.
Regarding the transition of driving from L3 to L4 robot taxis, Rivian’s Abdullah Zaidi showed the slide below outlining the different needs for cameras, radars, and lidars, and also the compute requirements.
Source: Rivian, AutoSens 2023, Detroit
No wonder automotive is such an attractive space for semiconductors. Computing, sensing, and data transport requirements are just growing enormously. And mind you that the picture above does not mention other cameras for in-cabin monitoring.
Besides the computing requirements, data transport is core to my day-to-day work. In one of his slides, Mercedes-Benz AG’s Konstantin Fichtner presented that the DrivePilot system records 33.73 GB of trigger measurements per minute – 281 times as much as it takes to watch a Netflix 4K stream. That’s a lot of data to transport across networks-on-chips (NoCs), between chips and chiplets. And it, of course, raises the question of on-board vs. off-board processing.
Are we there yet? Not quite, but we are getting closer. On the last day of the conference, Carnegie Mellon University’s Prof. Philip Koopman sobered up the audience with his talk “Defining Safety For Shared Human/Computer Driver Responsibility.” His keynote walked the audience through the accountability dilemma when a highly automated vehicle crashes and made some constructive suggestions on updating state laws. In their recently published essay “Winning the Imitation Game: Setting Safety Expectations for Automated Vehicles.” Prof. Koopman and William H. Widen from the University of Miami School of Law suggest that legislatures amend existing laws to create a new legal category of “computer driver” to allow a plaintiff to make a negligence claim.
To make that exact point, the universe created the situation the next day, which I took a picture of below. Can you see what’s wrong here?
Source: Frank Schirrmeister, May 2023
Yep, a pedestrian ghost in the machine.
To technology’s excuse, there had been a jay-walking pedestrian about 30 seconds ago, which probably erred on the side of caution. But still, this was a good reminder that future sensors are hopefully better than my eyes, and a thermal sensor would have helped here too.
All soberness and glitches aside, let’s not forget the end goal: Reducing fatalities due to traffic situations. And as I joked in my last blog on ISO 26262 safety, How Safe is Safe Enough: “If aliens would arrive and assess how to reduce traffic-related deaths, they would most certainly take humans off the streets.”
Brave new autonomous world, here we come. And I am keeping my license. That 1997 Miata doesn’t drive itself!
Fullmoonfever
Top 20
Been wondering about the Renesas tape out and also BRN comments on communications opportunities.
I see that Renesas acquired Celeno Communications out of Israel in Dec 21.
They also acquired Reality AI who is strong in AI algos apparently and Renesas also obviously have a licence and tech partnership with us for SNN.
Just musing if our opp with Renesas could also dovetail into someone like Celeno as a subsidiary with the focus on AIoT and IIoT.
Would keep it all an essentially in house working group
Renesas actually have a bit going on entering into the auto radar mkt as well.
semiengineering.com
AIoT is driving a shift from centralized, cloud-based architectures to distributed, edge-based designs.
JANUARY 26TH, 2023 - BY: SAILESH CHITTIPEDDI
Last year, I reflected on the Renesas Renaissance in terms of how our long-term growth strategy is positioning the company as a full-spectrum, global technology solutions provider with an extended physical footprint in the U.S., Europe, and China. Thanks to the acquisitions of Intersil, IDT, Dialog Semiconductor, and Celeno we now have expansive design capabilities that surround our embedded processor expertise with four core analog and mixed-signal competencies: sensors and sensor signal conditioning, connectivity, actuation and power management.
In each case, these companies and their engineers and scientists are contributing significantly to our key growth objectives, with overall revenue increasing at a CAGR of 17.7% between 2019 and 2021 and operating margin growing 16.9% over the same period. Perhaps most telling is the fact that, between 2020 and 2022, our infrastructure and industrial businesses grew by 33% and 37%, respectively, while our internet of things (IoT) segment saw an astounding 79% jump (37% organically) in CAGR.
That the industrial IoT (IIoT) marketplace is expanding at such a torrid pace is not surprising. For Renesas it signals an opportunity to accelerate adoption among our customers and ecosystem partners by enabling the convergence of three key technology areas that are maturing at roughly the same time: IoT, 5G connectivity and artificial intelligence (AI). We call this AI IoT, or AIoT, and the trend is driving a shift in how we collect, store, process, distribute, secure and power data in order to turn it into actionable intelligence we can learn from.
Such a sea change entails a move away from centralized, cloud-based architectures to distributed, edge-based designs that use tiny machine learning (ML) nodes like MCUs and MPUs to define the endpoint, accelerate mathematical models and improve the performance of deep neural networks.
While many people might associate AI with futuristic consumer applications like robotic assistants, the fact is that much of the initial impact will be felt in the industrial space where when IoT endpoint node creation is exploding at an 85% CAGR (2017-2025), yielding an almost unfathomable 73 zetabytes of data, according to IDC. From an applications perspective, these growth lines will open new markets and revenue channels in areas such as predictive maintenance, rapid defect detection, biometric recognition, and asset tracking, to name just a few.
That’s what led us to one of our most recent and significant 2022 acquisitions – Reality AI. The company is especially strong in developing algorithms for the industrial space, which is helping to fulfill our long-term vision of combining advanced signal processing and mathematical modeling with AI to build machine learning models that we can implement on our embedded processors – from 16 to 64 bits.
The Reality AI acquisition is an important component of realizing our AIoT vision, which this year also included investments in companies like Syntiant and Arduino that are extending our reach into more complex use cases, as well as a platinum sponsorship with the Tiny ML Consortium.
For Renesas, those strategic investments are part of a long-term, three-pronged approach to enabling AIoT, which includes access to world-class MCU and MPU devices and a commitment to the broader AIoT ecosystem, where we have more than 200 technology partners. Together with 5G and other forms of wireless connectivity, such as WiFi 6/6E/7 and near-field communications (NFC), we are enabling a fully distributed AIoT network that will revolutionize how the industrial workplace is managed by bringing constant, on-device learning and decision making to the factory floor.
I see that Renesas acquired Celeno Communications out of Israel in Dec 21.
They also acquired Reality AI who is strong in AI algos apparently and Renesas also obviously have a licence and tech partnership with us for SNN.
Just musing if our opp with Renesas could also dovetail into someone like Celeno as a subsidiary with the focus on AIoT and IIoT.
Would keep it all an essentially in house working group
Renesas actually have a bit going on entering into the auto radar mkt as well.
The New Convergence: IoT, AI, And 5G Bring Actionable Intelligence To The Factory Floor
AIoT is driving a shift from centralized, cloud-based architectures to distributed, edge-based designs.
semiengineering.com
The New Convergence: IoT, AI, And 5G Bring Actionable Intelligence To The Factory Floor
AIoT is driving a shift from centralized, cloud-based architectures to distributed, edge-based designs.
JANUARY 26TH, 2023 - BY: SAILESH CHITTIPEDDI
Last year, I reflected on the Renesas Renaissance in terms of how our long-term growth strategy is positioning the company as a full-spectrum, global technology solutions provider with an extended physical footprint in the U.S., Europe, and China. Thanks to the acquisitions of Intersil, IDT, Dialog Semiconductor, and Celeno we now have expansive design capabilities that surround our embedded processor expertise with four core analog and mixed-signal competencies: sensors and sensor signal conditioning, connectivity, actuation and power management.
In each case, these companies and their engineers and scientists are contributing significantly to our key growth objectives, with overall revenue increasing at a CAGR of 17.7% between 2019 and 2021 and operating margin growing 16.9% over the same period. Perhaps most telling is the fact that, between 2020 and 2022, our infrastructure and industrial businesses grew by 33% and 37%, respectively, while our internet of things (IoT) segment saw an astounding 79% jump (37% organically) in CAGR.
That the industrial IoT (IIoT) marketplace is expanding at such a torrid pace is not surprising. For Renesas it signals an opportunity to accelerate adoption among our customers and ecosystem partners by enabling the convergence of three key technology areas that are maturing at roughly the same time: IoT, 5G connectivity and artificial intelligence (AI). We call this AI IoT, or AIoT, and the trend is driving a shift in how we collect, store, process, distribute, secure and power data in order to turn it into actionable intelligence we can learn from.
Such a sea change entails a move away from centralized, cloud-based architectures to distributed, edge-based designs that use tiny machine learning (ML) nodes like MCUs and MPUs to define the endpoint, accelerate mathematical models and improve the performance of deep neural networks.
While many people might associate AI with futuristic consumer applications like robotic assistants, the fact is that much of the initial impact will be felt in the industrial space where when IoT endpoint node creation is exploding at an 85% CAGR (2017-2025), yielding an almost unfathomable 73 zetabytes of data, according to IDC. From an applications perspective, these growth lines will open new markets and revenue channels in areas such as predictive maintenance, rapid defect detection, biometric recognition, and asset tracking, to name just a few.
That’s what led us to one of our most recent and significant 2022 acquisitions – Reality AI. The company is especially strong in developing algorithms for the industrial space, which is helping to fulfill our long-term vision of combining advanced signal processing and mathematical modeling with AI to build machine learning models that we can implement on our embedded processors – from 16 to 64 bits.
The Reality AI acquisition is an important component of realizing our AIoT vision, which this year also included investments in companies like Syntiant and Arduino that are extending our reach into more complex use cases, as well as a platinum sponsorship with the Tiny ML Consortium.
For Renesas, those strategic investments are part of a long-term, three-pronged approach to enabling AIoT, which includes access to world-class MCU and MPU devices and a commitment to the broader AIoT ecosystem, where we have more than 200 technology partners. Together with 5G and other forms of wireless connectivity, such as WiFi 6/6E/7 and near-field communications (NFC), we are enabling a fully distributed AIoT network that will revolutionize how the industrial workplace is managed by bringing constant, on-device learning and decision making to the factory floor.
Fullmoonfever
Top 20
Sensors Converge happening.at the mo.
www.sensorsconverge.com
I see Prophesee and Oculi (recall that name) speaking.
But also exhibiting are Renesas and Socionext who.
At this year’s event, Socionext will showcase its new automotive radar sensor technology including advanced RF CMOS Sensors for in-vehicle driver and passenger monitoring systems deliver ground-breaking functional and safety benefits
Sensors Converge
www.sensorsconverge.com
I see Prophesee and Oculi (recall that name) speaking.
But also exhibiting are Renesas and Socionext who.
At this year’s event, Socionext will showcase its new automotive radar sensor technology including advanced RF CMOS Sensors for in-vehicle driver and passenger monitoring systems deliver ground-breaking functional and safety benefits
FlipDollar
Never dog the boys
This thread has become more about every other company, than it is about BRN 
Still patiently waiting
happy hump day!
Still patiently waiting
happy hump day!
D
Deleted member 118
Guest
This page would be eerily quiet if they neverThis thread has become more about every other company, than it is about BRN
Still patiently waiting
Boab
I wish I could paint like Vincent
Podcast is up
brainchip.com
Episode 27: Podcast with USC Professor Dr. Gaurav Sukhatme on Innovation in AI, Robotics & Autonomy
Listen to Episode 27 of our podcast featuring USC Professor Dr. Gaurav Sukhatme as we explore innovation in AI, robotics, and autonomy.
brainchip.com
equanimous
Norse clairvoyant shapeshifter goddess
Brainchip is essentially for every other company to use to their advantage. 2024-2025 is looking like right time frame to shineThis thread has become more about every other company, than it is about BRN
Still patiently waiting
D
Deleted member 118
Guest
Drop in sp then.Podcast is up
Episode 27: Podcast with USC Professor Dr. Gaurav Sukhatme on Innovation in AI, Robotics & Autonomy
Listen to Episode 27 of our podcast featuring USC Professor Dr. Gaurav Sukhatme as we explore innovation in AI, robotics, and autonomy.
Similar threads
