Congratulations!
John Hopfield received the Nobel Prize in physics for the eponymous Hopfield network—the energy-based associative memory model, which has influenced the course of both neuroscience and machine learning as well as many other fields.
Nobel announcement

We published two papers at the IEEE High Performance Extreme Computing conference on NorthPole.

• We reported NorthPole running an LLM for the first time. See the IBM Research Blog
• We described our new rugged, embedded VPX form-factor NorthPole board for edge applications.

NorthPole updates over the last few months

• NorthPole inducted into the Computer History Museum link
  
• I presented to students of the Dwight D. Eisenhower School for National Security and Resource Strategy: link
  
• NorthPole's energy efficiency mention by the BBC and the Economist

I was invited to give a NorthPole talk at the Neuro Inspired Computational Elements (NICE) Conference at Scripps Institution of Oceanography. Neuroscience and neuromorphic researchers from around the globe converged for a couple of days near San Diego. Carver Mead was recognized for his lifetime contribution to neuromorphic engineering with a special Misha Mahowald Prize, named after his student with whom he founded the field.
It's always good to catch up with neuromorphic colleagues from the past as well as to meet some new ones.

We were invited to submit a NorthPole paper, give a talk (given by Andrew), and demonstrate it running live for the first time (showing the work from SW and HW colleagues). It was my first time at ISSCC, and it was great to speak with so many fellow chip designers.

NorthPole was featured in The 2023 IBM Research Annual Letter in the "What's next in semiconductors section", highlighting its Neuro-inspired origin and architectural innovations to eliminate the von Neumann bottleneck, which significantly improves energy and space efficiency versus existing chips.

The 2023 IBM Research Annual Letter

NorthPole has received a lot of attention from the press as well as an IBM Research blog post. Many of the articles are insightful, well written, and factual. Here are a few of the best ones:

• Forbes
• Nature News
• IEEE Spectrum
• Ars Technica
• IBM Research blog

There is also a great deal of information at modha.org.

Our NorthPole Chip was published in Science and described in a featured story. NorthPole is a high-performance, high-efficiency neural network inference accelerator. It uses local memory within a parallel, distributed array of 256 cores, linked by NoCs. Fabbed in a 12nm finFET process, the chip has 224 MB of SRAM and uses 22 billion transistors. It represents the most efficient chip to date for several inference tasks such as Resnet-50 classification and Yolov4 detection.
Science (paywall).

It's been a while since I've posted. Our NorthPole Chip was publicly presented for the first time at Hot Chips on 29 August 2023. NorthPole represents many years of intensive work by many people and we are happy to announce that it is fully functional in first silicon. The slides and video will be available at hotchips.org. Please see modha.org as well.

I returned to DAC this year in San Franscico, chairing a session on Industrial-Strength Accelerators for Machine Learning and Artificial Intelligence.

I attended last year in Austin as one of the inaugural Five Outstanding Innovators Under 40.

IBM Research's Brain-Inspired Computing Group is growing our team again. We are focusing our search on hardware engineers to build the next generation of deep neural network hardware. In our last generation chip, we created a new architecture, designed and fabricated the chip, and built an ecosystem around it, including algorithms and development tools. Available positions for the next generation:

• Hardware Research Scientists
We are looking for several hardware researchers for all aspects of deep neural network hardware. Apply here
• Logic Designers and Verification Engineers
We are looking for several logic designers and verification engineers to contribute to creation and verification of deep neural network hardware. Apply here or here

Please feel free to contact me with questions or just to let me know you applied at arthurjo@us.ibm.com or through Linkedin.

For much more information and more positions go here

We are growing our team. If you are a knowledgeable hardware engineer with an interest in machine learning or a knowledgeable machine learning researcher with an interest in specialized hardware, please contact me at john@johnarthur.org and apply here.

In the last year, we published two papers on training algorithms for TrueNorth. Our algorithms are among the first to focus on low precision deep neural networks and are the first to train networks with full constraints for specialized hardware and deploy networks on hardware. In addition, we submitted two preprints to arxiv in the same domain:

• S.K. Esser, P.A. Merolla, J.V. Arthur, et al (2016). "Convolutional Networks for Fast, Energy-Efficient Neuromorphic Computing". PNAS 113 (41) 11441-11446 (arxiv preprint).
see blog post by author Steve Esser here
• S.K. Esser, R. Appuswamy, P. Merolla, J.V. Arthur, D.S. Modha (2015). "Backpropagation for energy-efficient neuromorphic computing." Adv. Neural Inf. Process. Syst. 28 (NIPS). (spotlight presentation), ranked among top ~5% of submissions
see blog post by author Steve Esser et al here
• R. Appuswamy, T. Nayak, J.V. Arthur, et al (2016). Structured Convolution Matrices for Energy-efficient Deep learning. arXiv:1606.02407.
• P. Merolla, R. Appuswamy, J.V. Arthur, S.K. Esser, and D.S. Modha (2016). Deep neural networks are robust to weight binarization and other non-linear distortions. arXiv:1606.01981.

TrueNorth was award a number of honors in the last year or so:

• 2015 R&D 100 Award as Editor's Choice in the IT/Electrical Category
• 2016 Inaugural Misha Mahowald Prize—This is particular meaningful as Misha was a founder of the neuromorphic engineering field and a great inspiration to me and many people that I respect
• 1st Place, IBM Pat Goldberg Math / CS / EE Best Paper Award for 2014
• TrueNorth was accepted in the Computer History Museum
linked here

Late last year we published a paper in Supercomputing describing the TrueNorth ecosystem and systems as well as applications developed using it by several researchers from a handful of institutions. We didn't just build a chip and then set it on a shelf; we built tools and made it usable. TrueNorth continues to be used by other researchers at events such as 2016 Telluride Neuromorphic Cognition Engineering Workshop and our bootcamp reunion as well as by industry partners.

In a short time, many researchers have been exposed to TrueNorth and developing neural networks for it.

• Three of my colleagues and I took TrueNorth to the Telluride Neuromorphic Cognition Engineering Workshop in July. Grad students, postdocs, and professors dove in and spent three weeks on intensive projects. Go here and here for more details.
• We organized a TrueNorth BootCamp for academic and government researchers to learn the basics of programming and running. Go here, here, here, and here for more.

Supercomputing 2014 takes place in New Orleans this week. Our paper was nominated as an ACM Gordon Bell Prize finalist.

We (the brain-inspired computing group) held the second colloquium at IBM Research - Almaden concerning cognitive computing. Our TrueNorth Chip as well as other neuromorphic systems featured prominently. More information can be found here and here.

I gave one of the first talks about the TrueNorth Chip and Architecture at HPEC `14, titled Computing with Neurosynaptic Cores

Our TrueNorth Chip was published in Science as well as depicted on the cover and featured in a news story. TrueNorth is the largest neuromorphic chip fabricated with one million configurable digital neurons and 256 million programmable synapse spread across 4,096 neurosynaptic cores, which only consume ~70mW. It has over 400Mbits of embedded SRAM and uses 5.4 billion transistors.
Science (paywall).

We are looking for hardware and software engineerings to develop the next generation brain-inspired computers and to develop algorithms and applications to run on them.
Jobs: Brain-inspired Computing.

The Stanford Neurogrid Chip was published in IEEE Proceedings in a special issue on "Engineering intelligent electronic systems based on computational neuroscience". Neurogrid is a system of 16 Neurocore Chips each with 65k analog neurons, interconnected by a tree router.
IEEE Proceedings (paywall).