Showing posts with label chips. Show all posts
Showing posts with label chips. Show all posts

Friday, August 2, 2024

Chips At Stake


Integrating dimensions to get more out of Moore's Law and advance electronics
Jan 2024, phys.org

Moore's Law - the number of transistors on a chip will double every two years (but a limit exists)

More Moore - vertically stacking multiple layers of semiconductor devices to beat the limit (3D integration)

More than Moore - transistors made from 2D materials (using 2D materials, not just 2D-like layers of 3D materials)

Monolithic Moore - monolithic 3D integration uses one brick of 2D things, instead of layers of 2D things, so there's no connections between the layers, so it saves space 

via Penn State: Darsith Jayachandran et al, Three-dimensional integration of two-dimensional field-effect transistors, Nature (2024). DOI: 10.1038/s41586-023-06860-5



Emulating neurodegeneration and aging in artificial intelligence systems
Apr 2024, phys.org

"We used IQ tests performed by large language models (LLMs) and, more specifically, the LLaMA 2, to introduce the concept of 'neural erosion. This deliberate erosion involves ablating synapses or neurons or adding Gaussian noise during or after training, resulting in a controlled decline in the LLMs' performance."

The researchers found that when they deliberately ablated (i.e., removed) some of the artificial synapses or neurons of the LLaMA 2 model, its performance on IQ tests declined, following a particular pattern.

"The LLM loses abstract thinking abilities, followed by mathematical degradation, and ultimately, a loss in linguistic ability, responding to prompts incoherently. ... We are now conducting further tests to better understand this observed pattern."

Interestingly, this 'neuro-erosion' pattern is aligned with the neurodegeneration patterns observed in humans.

via University of California Irvine: Antonios Alexos et al, Neural Erosion: Emulating Controlled Neurodegeneration and Aging in AI Systems, arXiv (2024). DOI: 10.48550/arxiv.2403.10596

Post Script: A new kind of chip called an NPU for neural processing unit, is like the new GPU, which was the new CPU:
Your current PC probably doesn’t have an AI processor, but your next one might
Feb 2024, Ars Technica

The comments in this article are the most clearly examined examples of what NPUs will do than any other thing I've read, and considering that the forum of Ars is known for the technical proficiency of its members: 
  • From the article, Intel Senior Director of Technical Marketing Robert Hallock  - "Camera segmentation, this whole background blurring thing... moving that to the NPU saves about 30 to 50 percent power versus running it elsewhere."
  • **RTS** - It is pretty shocking looking at powermetrics on macOS to see how rare the Neural Engine is even powered as most applications (even AI exclusive applications like stable diffusion/LLM frontends) just run any AI work they need on the CPU or GPU.
  • AlicePlaysWithRockets - I use it daily in Final Cut Pro and meeting camera effects.
  • LlamaDragon - NPUs can be used by Photoprism to speed up importing of images (it tries to ID pics with dogs or cars or "cooking" and so on) or by something like Frigate (camera monitoring) to do similar ID-ing in real time, it might be useful.
  • AmanoJyaku Ars Praefectus - [why] It has to do with CPUs being general purpose, and therefore able to do anything, vs. hardware accelerators that do specific things. Hardware accelerators can't do most of the things CPUs can do, but the things they can do can be done much faster than CPUs do them. In turn, this makes them more power efficient. The trend started with graphics cards, continued with sound cards and network cards, and have since grown to include other devices, some of which accelerate things as minute as image decoding. ... Theoretically, the most efficient device is one that has accelerators for everything you do on a device. However, there will always be new things, so CPUs are unlikely to be eliminated. As for what makes general-purpose CPUs different from task-specific accelerators, forum posts can't easily sum this up. That's a topic explained in college courses, and the basis for entire careers. 
  • autostop - The Neural Engine on the Apple chips is what makes "Live Text" (universal OCR) work on the Macintosh and iPhone. (You can open a scanned PDF, or a screen grab, and the mouse pointer turns into an I-beam and you can highlight and copy just as if it were text in Word or something.)
  • Siosphere - Users who view photos a lot on their laptop/desktop, it could be used for on-device people finding, or computing interoplation for zooming in/out of a photo so it is smoother/faster. ... It could be used for generic type-ahead suggestions, which if they suck is super annoying, but when they work well are a very useful timesaver. (When they actually suggest what I was actually going to write, I'm really pleased, when they suggest something else they are annoying, but they are getting better). ... Better searching capabilities by understanding more of what you are searching for on your computer, being smarter about intent, are you searching for a specific file by name, type of file, application, etc. ... Generalized automation of repetitive tasks, things you might currently script yourself could be automatically found and setup (again, this is annoying if it doesn't do what you want, but I'm just saying best case it automates in the exact way you are wanting). ... Faster windows hello (if you use that), better camera processing for video calls, better workload predicting for changing the task priority to save battery on laptops, or give CPU power more to the things that need it. ... There are a lot of subtle ways that a dedicated AI chip could be used, and there will be so many more when it is just an available resource any developers can tap into
  • Isaacc7 - NPUs can be used to accelerate many tasks like voice recognition, camera processing, live subtitles, etc. 
  • Toastr - Also things like on-device automatic image tagging of faces for your photo library, which is a big plus for me who wants to organize my family photo library but doesn't like the privacy implications of cloud photo services like Google Photos. ... Or, for another example:
  • The self-hosted Frigate NVR software supports object detection via CPU, GPU, or NPU. Sure, you could run it on an RTX 4080, but you could also just throw a $25 Coral Accelerator in a mini PC and get plenty of performance but with a device that draws just a few watts.
  • longhornchris04 - Side note, while GPUs can do the highly parallel low precision computing, they are designed for graphics which generally require a higher level of precision. So yes, GPUs can do the work, and do it quite easily, but they are often overkill for the job and thus are less efficient at it. 
  • or just anandtech: https://www.anandtech.com/show/20046/intel-unveils-meteor-lake-architecture-intel-4-heralds-the-disaggregated-future-of-mobile-cpus/4

Posted by at No comments:
Labels: , , , , ,

Wednesday, July 10, 2024

Everything Is A Computer If You Try Hard Enough


You walk into a room, and it's empty, but you enter, and without hesitation, you sit down into mid air, and before you land, a chair materializes itself to catch you. The actual matter, the particles that make up the chair, have been engineered to become chairs. They have "chair" written not into their DNA, but into their fundamental physics. They have no power source because they take their energy from light waves, sound waves, vibrations, even gradients like in between high temperatures and lows, or saltwater and fresh, or whatever that means, since it could be an information gradient (could it?). They have no battery because they don't use more than they need in real time. Every particle of this special form of matter is a computer, with wireless communication, with sensors, all of it built into the physics of the particles themselves. It will know you're in the room, who you are and what you want. And that's why all you have to do is want to sit, and the chair materializes. It's still too hard to explain what it means when everything is a computer, but here is an introduction:  

Photonic chip that 'fits together like Lego' opens door to semiconductor industry
Dec 2023, phys.org

Chiplets - The chip is built using an emerging technology in silicon photonics that allows the integration of diverse systems on semiconductors less than 5 millimeters wide; it's like fitting together Lego building blocks, where new materials are integrated through advanced packaging of components, using electronic "chiplets."

via University of Sydney Nano Institute:  Matthew Garrett et al, Integrated microwave photonic notch filter using a heterogeneously integrated Brillouin and active-silicon photonic circuit, Nature Communications (2023). DOI: 10.1038/s41467-023-43404-x

Unexpected AI fingers image credit: AI Art - AI Fingers at Work on a Circuitboard - 2023


Study suggests that physical processes can have hidden neural network-like abilities
Jan 2024, phys.org

Natural molecular processes can do complex calculations that rival a simple neural network. On the face of it, the initial steps in the act of freezing - called 'nucleation' in physics - do not resemble 'thinking'. But the new study shows that the act of freezing can "recognize" subtly different chemical combinations - e.g., the smell of oatmeal raisin cookies versus chocolate chip - and build different molecular structures in response.

The work points at a new view of computation that does not involve designing circuits, but rather designing what physicists call a phase diagram. For example, for water, a phase diagram might describe the temperature and pressure conditions in which liquid water will freeze or boil, which are 'muscle'-like material properties. But this work shows that the phase diagram can also encode 'thinking' in addition to 'doing,' when scaled up to complex systems with many different kinds of components.

via University of Chicago, California Institute of Technology, and Maynooth University: Constantine Glen Evans et al, Pattern recognition in the nucleation kinetics of non-equilibrium self-assembly, Nature (2024). DOI: 10.1038/s41586-023-06890-z

 
International research team develops new hardware for neuromorphic computing
Feb 2024, phys.org

In the eye, the visual information is pre-processed by hundreds of millions of the retina's photoreceptors and converted into electrical signals that are transmitted by the optic nerve to the brain. This process greatly reduces the amount of data processed in the brain by the visual cortex.

(The nose is far, far crazier in what it does, by the way)

Inspired by eyesight, this on-chip phonon-magnon reservoir for neuromorphic computing maps input signals into a multidimensional reservoir space. The reservoir is made of acoustic waves (phonons) and spin waves (magnons), and is not trained but only expedites recognition by a simplified artificial neural network, resulting in enormous reduction of computational resources and training time.

(Wait until they get inspired by the nose)

via Technische Universität Dortmund, Loughborough University, V. E. Lashkaryov Institute of Semiconductor Physics in Kyiv, University of Nottingham: Dmytro D. Yaremkevich et al, On-chip phonon-magnon reservoir for neuromorphic computing, Nature Communications (2023). DOI: 10.1038/s41467-023-43891-y


A low-cost system to collect EEG measurements during VR experiences
Feb 2024, phys.org

Everything gets it's own chip, just like this

NeuroVista, the new system proposed by the researchers, utilizes KS1092, a cost-effective biological potential measurement chip. The prototype of the device created by the researchers is comprised of this chip, along with a set of electrodes, and a lithium battery.

via South China University of Technology: Zhiyuan Yu et al, A low-cost, wireless, 4-channel EEG measurement system used in virtual reality environments, HardwareX (2024). DOI: 10.1016/j.ohx.2024.e00507.
 

Giant leap toward neuromorphic devices: High-performance spin-wave reservoir computing
Mar 2024, phys.org

It's a high-performance spin wave reservoir computing that uses spintronics. It works with a randomly generated network called the "reservoir" which enables the memorization of past input information and its nonlinear transformation, allowing physical systems to perform tasks for sequential data.

via Tohoku University Advanced Institute for Materials Research: Satoshi Iihama et al, Universal scaling between wave speed and size enables nanoscale high-performance reservoir computing based on propagating spin-waves, npj Spintronics (2024). DOI: 10.1038/s44306-024-00008-5

Posted by at No comments:
Labels: , , , , , , , ,

Wednesday, January 17, 2024

Vision Technologies See Way Ahead


Neuromorphic camera and machine learning aid nanoscopic imaging
Feb 2023, phys.org

Brain-inspired image sensor using machine learning can go beyond the diffraction limit of light to detect minuscule objects such as cellular components or nanoparticles smaller than 50 nanometers in size, and invisible to current microscopes.

via Indian Institute of Science: Rohit Mangalwedhekar et al, Achieving nanoscale precision using neuromorphic localization microscopy, Nature Nanotechnology (2023). DOI: 10.1038/s41565-022-01291-1



New 'camera' with shutter speed of 1 trillionth of a second sees through dynamic disorder of atoms
Mar 2023, phys.org

Doesn't work like a conventional camera - it uses neutrons from a source at the U.S. Department of Energy's Oak Ridge National Laboratory (ORNL) to measure atomic positions with a shutter speed of around one picosecond, or a million million (a trillion) times faster than normal camera shutters. 

via Columbia University School of Engineering and Applied Science: Simon A. J. Kimber et al, Dynamic crystallography reveals spontaneous anisotropy in cubic GeTe, Nature Materials (2023). DOI: 10.1038/s41563-023-01483-7


Superconducting nanowire camera will explore brain cells, space
Jul 2023, phys.org

Superconducting camera -- A pixel array 400 times greater than previous largest photon camera, this is a 400,000 pixel superconducting nanowire single-photon detector (SNSPD), for light frequencies from the visible to ultraviolet and infrared range and speed rates in the picoseconds.

via National Institute of Standards and Technology in Boulder, University of Colorado's Department of Physics and the Jet Propulsion Laboratory at the California Institute of Technology: Bakhrom G. Oripov et al, A superconducting-nanowire single-photon camera with 400,000 pixels, arXiv (2023). DOI: 10.48550/arxiv.2306.09473

AI Art - Skeleton Reading an X-Ray - 2024

The first network of robotic telescopes present across five continents is deployed
Feb 2023, phys.org

The existence of a network of very fast pointing robotic telescopes such as BOOTES represents an ideal complement to satellite detection and, in fact, BOOTES will also work to track and monitor neutrino sources and objects that emit gravitational waves, or even objects such as comets, asteroids, variable stars or supernovae. But it will also keep an eye on the sky, both in tracking space debris and potentially dangerous objects that may pose a threat to our planet.

via Spanish National Research Council: Youdong Hu et al, The Burst Observer and Optical Transient Exploring System in the multi-messenger astronomy era, Frontiers in Astronomy and Space Sciences (2023). DOI: 10.3389/fspas.2023.952887.


HotSat-1: Spacecraft to map UK's heat inefficient buildings
Jun 2023, BBC News

Mass surveillance -- 

At an altitude of 500km (311 miles), infrared satellite HotSat-1, funded by the UK and European space agencies, manufactured by Surrey Satellite Technology Ltd in Guildford, and to be operated by the London-based start-up Satellite Vu, will identify dwellings wasting energy.

The data will also provide intelligence to the financial and insurance sectors - and even the military - by showing how temperatures in a scene change over time. It's possible, for example, to get a sense of the volume and type of output from a factory just from its heat signature.

Pollution monitoring ought to be another application. Watching for sudden changes in the temperature of river water might be an indicator that something is awry.

Update: A novel UK satellite has returned its first pictures of heat variations across the surface of the Earth. HotSat-1: UK spacecraft maps heat variations across Earth, Sep 2023, BBC News


The future of AI hardware: Scientists unveil all-analog photoelectronic chip
Oct 2023, phys.org

All-analog photoelectronic chip that combines optical and electronic computing. It's specifically for visual data processing (as expected). 

New words to me - "diffractive neural network"
Also - ACCEL: all-analog chip combining electronic and light computing

via Tsinghua University: Yitong Chen et al, All-analog photoelectronic chip for high-speed vision tasks, Nature (2023). DOI: 10.1038/s41586-023-06558-8

Also: Computer vision accelerated using photons and electrons, Nature (2023). DOI: 10.1038/d41586-023-02947-1 


Posted by at No comments:
Labels: , , , ,

Wednesday, January 10, 2024

Make Everything a Computer Again


AKA The Atoms Themselves Are Computers Part 2

Tiny device mimics human vision and memory abilities
Jun 2023, phys.org

There won't be any computers one day. Somehow things will compute by themselves because of the way they're designed. Each thing will compute differently because it will be made of different things and arranged in different ways. There won't be all-purpose computers anymore; some things will see, some will hear, some will count, maybe some will smell.

A neuromorphic vision device -- a single chip enabled by a sensing element, doped indium oxide,  thousands of times thinner than a human hair and requires no external parts to operate, captures, processes and stores visual information.

The device mimics a human eye's ability to capture light, pre-packages and transmits information like an optical nerve, and stores and classifies it in a memory system like the way our brains can.

via Royal Melbourne Institute of Technology RMIT, Deakin University and University of Melbourne: Aishani Mazumder et al, Long Duration Persistent Photocurrent in 3 nm Thin Doped Indium Oxide for Integrated Light Sensing and In‐Sensor Neuromorphic Computation, Advanced Functional Materials (2023). DOI: 10.1002/adfm.202303641



Physicists design metamaterials with built-in frustration for mechanical memory
Jun 2023, phys.org

The future of computing where everything is a computer -- "metamaterials are materials whose responses are determined by their structure rather than their chemical composition"

But this metamaterial now has memory -- To construct a metamaterial with mechanical memory, they realized that its design needs to be "frustrated," and that this frustration corresponds to a new type of order, which they call non-orientable order. These materials naturally want to be ordered, but something in their structure forbids the order to span the whole system and forces the ordered pattern to vanish at one point or line in space. There is no way to get rid of that vanishing point without cutting the structure, so it has to be there no matter what.

(A simple example of a non-orientable object is a Möbius strip)

via University of Amsterdam: Xiaofei Guo, Non-orientable order and non-commutative response in frustrated metamate, Nature (2023). DOI: 10.1038/s41586-023-06022-7


New type of computer memory could greatly reduce energy use and improve performance
Jun 2023, phys.org

Processes data in a similar way as the synapses in the human brain, and based on hafnium oxide.

"In conventional computing, there's memory on one side and processing on the other, and data is shuffled back between the two, which takes both energy and time."

Conventional memory devices are capable of two states: one or zero. A functioning resistive switching memory device however, would be capable of a continuous range of states

At the atomic level, hafnium oxide has no structure, with the hafnium and oxygen atoms randomly mixed, making it challenging to use for memory applications.

However, the researchers found that by adding barium to thin films of hafnium oxide, some unusual structures started to form (and which allow electrons to pass through), perpendicular to the hafnium oxide plane, in the composite material.

via University of Cambridge: Markus Hellenbrand et al, Thin-film design of amorphous hafnium oxide nanocomposites enabling strong interfacial resistive switching uniformity, Science Advances (2023). DOI: 10.1126/sciadv.adg1946

AI Art - Nanotechnology Activated by a Frequency in the Human Body - 2023

The catch-22s of reservoir computing: Researchers find overlooked weakness in powerful machine learning tool
Sep 2023, phys.org

Sante Fe Institute is always far out:

Reservoir computing is effective in predicting the trajectory of chaotic systems after seeing very little training data, and can even determine where the system would end up just from its initial conditions.

"In a sense, you have this kind of information sneaked in before the training begins," he says. And if they perturbed the model? "Generally, it performed really poorly," Zhang says. That suggests that the model cannot make accurate predictions unless key information about the system being predicted was already built in. For RC, the duo observed that in order to correctly predict the system, the model requires a lengthy "warm-up" time that's almost as time-consuming as the dynamic movements of the magnet itself.

via Santa Fe Institute and Toronto Metropolitan University: Yuanzhao Zhang et al, Catch-22s of reservoir computing, Physical Review Research (2023). DOI: 10.1103/PhysRevResearch.5.033213


New 'assembly theory' unifies physics and biology to explain evolution and complexity
Oct 2023, phys.org

Assembly Theory - developing as empirically validated approach to life detection, with implications for the search for alien life and efforts to evolve new life forms in the laboratory.

In prior work, the team assigned a complexity score to molecules called the molecular assembly index, based on the minimal number of bond-forming steps required to build a molecule. They showed how this index is experimentally measurable and how high values correlate with life-derived molecules.

The new study introduces mathematical formalism around a physical quantity called "assembly" that captures how much selection is required to produce a given set of complex objects, based on their abundance and assembly indices.

"Assembly theory provides a completely new lens for looking at physics, chemistry and biology as different perspectives of the same underlying reality," explained lead author Professor Sara Walker, a theoretical physicist and origin of life researcher from Arizona State University.

via University of Glasgow and Arizona State University: Leroy Cronin, Assembly theory explains and quantifies selection and evolution, Nature (2023). DOI: 10.1038/s41586-023-06600-9. 

Posted by at No comments:
Labels: , , , ,
Subscribe to: Posts (Atom)