University of Sussex academics have established a method of turbocharging desktop PCs to give them the same capability as supercomputers worth tens of millions of pounds.
Dr James Knight and Prof Thomas Nowotny from the University of Sussex’s School of Engineering and Informatics used the latest Graphical Processing Units (GPUs) to give a single desktop PC the capacity to simulate brain models of almost unlimited size.
The researchers believe the innovation, detailed in Nature Computational Science, will make it possible for many more researchers around the world to carry out research on large-scale brain simulation, including the investigation of neurological disorders.
Currently, the cost of supercomputers is so prohibitive they are only affordable to very large institutions and government agencies and so are not accessible for large numbers of researchers.
As well as shaving tens of millions of pounds off the costs of a supercomputer, the simulations run on the desktop PC require approximately 10 times less energy bringing a significant sustainability benefit too.
Dr Knight, Research Fellow in Computer Science at the University of Sussex, said: “I think the main benefit of our research is one of accessibility. Outside of these very large organisations, academics typically have to apply to get even limited time on a supercomputer for a particular scientific purpose. This is quite a high barrier for entry which is potentially holding back a lot of significant research.
“Our hope for our own research now is to apply these techniques to brain-inspired machine learning so that we can help solve problems that biological brains excel at but which are currently beyond simulations.
“As well as the advances we have demonstrated in procedural connectivity in the context of GPU hardware, we also believe that there is also potential for developing new types of neuromorphic hardware built from the ground up for procedural connectivity. Key components could be implemented directly in hardware which could lead to even more truly significant compute time improvements.”
The research builds on the pioneering work of US researcher Eugene Izhikevich who pioneered a similar method for large-scale brain simulation in 2006.
At the time, computers were too slow for the method to be widely applicable meaning simulating large-scale brain models has until now only been possible for a minority of researchers privileged to have access to supercomputer systems.
The researchers applied Izhikevich’s technique to a modern GPU, with approximately 2,000 times the computing power available 15 years ago, to create a cutting-edge model of a Macaque’s visual cortex (with 4.13 × 106 neurons and 24.2 × 109 synapse) which previously could only be simulated on a supercomputer.
The researchers’ GPU accelerated spiking neural network simulator uses the large amount of computational power available on a GPU to ‘procedurally’ generate connectivity and synaptic weights ‘on the go’ as spikes are triggered – removing the need to store connectivity data in memory.
Initialization of the researchers’ model took six minutes and simulation of each biological second took 7.7 min in the ground state and 8.4 min in the resting state- up to 35 % less time than a previous supercomputer simulation. In 2018, one rack of an IBM Blue Gene/Q supercomputer initialization of the model took around five minutes and simulating one second of biological time took approximately 12 minutes.
Prof Nowotny, Professor of Informatics at the University of Sussex, said: “Large-scale simulations of spiking neural network models are an important tool for improving our understanding of the dynamics and ultimately the function of brains. However, even small mammals such as mice have on the order of 1 × 1012 synaptic connections meaning that simulations require several terabytes of data – an unrealistic memory requirement for a single desktop machine.
“This research is a game-changer for computational Neuroscience and AI researchers who can now simulate brain circuits on their local workstations, but it also allows people outside academia to turn their gaming PC into a supercomputer and run large neural networks.”
My first quantum brain posting! (Well, I do have something that seems loosely related in a July 5, 2017 posting about quantum entanglement and machine learning and more. Also, I have lots of item on brainlike or neuromorphic computing.)
Getting to the latest news, a February 1, 2021 news item on Nanowerk announces research in to new intelligent materials that could lead to a ‘quantum brain’,
An intelligent material that learns by physically changing itself, similar to how the human brain works, could be the foundation of a completely new generation of computers. Radboud [university in the Netherlands] physicists working toward this so-called “quantum brain” have made an important step. They have demonstrated that they can pattern and interconnect a network of single atoms, and mimic the autonomous behaviour of neurons and synapses in a brain.
If I understand the difference between the work in 2017 and this latest work, it’s that in 2017 they were looking at quantum states and their possible effect on machine learning, while this work in 2021 is focused on a new material with some special characteristics.
Considering the growing global demand for computing capacity, more and more data centres are necessary, all of which leave an ever-expanding energy footprint. ‘It is clear that we have to find new strategies to store and process information in an energy efficient way’, says project leader Alexander Khajetoorians, Professor of Scanning Probe Microscopy at Radboud University.
‘This requires not only improvements to technology, but also fundamental research in game changing approaches. Our new idea of building a ‘quantum brain’ based on the quantum properties of materials could be the basis for a future solution for applications in artificial intelligence.’
For artificial intelligence to work, a computer needs to be able to recognise patterns in the world and learn new ones. Today’s computers do this via machine learning software that controls the storage and processing of information on a separate computer hard drive. ‘Until now, this technology, which is based on a century-old paradigm, worked sufficiently. However, in the end, it is a very energy-inefficient process’, says co-author Bert Kappen, Professor of Neural networks and machine intelligence.
The physicists at Radboud University researched whether a piece of hardware could do the same, without the need of software. They discovered that by constructing a network of cobalt atoms on black phosphorus they were able to build a material that stores and processes information in similar ways to the brain, and, even more surprisingly, adapts itself.
In 2018, Khajetoorians and collaborators showed that it is possible to store information in the state of a single cobalt atom. By applying a voltage to the atom, they could induce “firing”, where the atom shuttles between a value of 0 and 1 randomly, much like one neuron. They have now discovered a way to create tailored ensembles of these atoms, and found that the firing behaviour of these ensembles mimics the behaviour of a brain-like model used in artificial intelligence.
In addition to observing the behaviour of spiking neurons, they were able to create the smallest synapse known to date. Unknowingly, they observed that these ensembles had an inherent adaptive property: their synapses changed their behaviour depending on what input they “saw”. ‘When stimulating the material over a longer period of time with a certain voltage, we were very surprised to see that the synapses actually changed. The material adapted its reaction based on the external stimuli that it received. It learned by itself’, says Khajetoorians.
Exploring and developing the quantum brain
The researchers now plan to scale up the system and build a larger network of atoms, as well as dive into new “quantum” materials that can be used. Also, they need to understand why the atom network behaves as it does. ‘We are at a state where we can start to relate fundamental physics to concepts in biology, like memory and learning’, says Khajetoorians.
If we could eventually construct a real machine from this material, we would be able to build self-learning computing devices that are more energy efficient and smaller than today’s computers. Yet, only when we understand how it works – and that is still a mystery – will we be able to tune its behaviour and start developing it into a technology. It is a very exciting time.’
Here is a charming image illustrating the reasons for a quantum brain,
Here’s a link to and a citation for the paper,
An atomic Boltzmann machine capable of self-adaption by Brian Kiraly, Elze J. Knol, Werner M. J. van Weerdenburg, Hilbert J. Kappen & Alexander A. Khajetoorians. Nature Nanotechnology (2021) DOI: https://doi.org/10.1038/s41565-020-00838-4 Published: 01 February 2021
An Oct. 29, 2020 news item on ScienceDaily features an explanation of the reasons for investigating brainlike (neuromorphic) computing ,
As progress in traditional computing slows, new forms of computing are coming to the forefront. At Penn State, a team of engineers is attempting to pioneer a type of computing that mimics the efficiency of the brain’s neural networks while exploiting the brain’s analog nature.
Modern computing is digital, made up of two states, on-off or one and zero. An analog computer, like the brain, has many possible states. It is the difference between flipping a light switch on or off and turning a dimmer switch to varying amounts of lighting.
Neuromorphic or brain-inspired computing has been studied for more than 40 years, according to Saptarshi Das, the team leader and Penn State [Pennsylvania State University] assistant professor of engineering science and mechanics. What’s new is that as the limits of digital computing have been reached, the need for high-speed image processing, for instance for self-driving cars, has grown. The rise of big data, which requires types of pattern recognition for which the brain architecture is particularly well suited, is another driver in the pursuit of neuromorphic computing.
“We have powerful computers, no doubt about that, the problem is you have to store the memory in one place and do the computing somewhere else,” Das said.
The shuttling of this data from memory to logic and back again takes a lot of energy and slows the speed of computing. In addition, this computer architecture requires a lot of space. If the computation and memory storage could be located in the same space, this bottleneck could be eliminated.
“We are creating artificial neural networks, which seek to emulate the energy and area efficiencies of the brain,” explained Thomas Shranghamer, a doctoral student in the Das group and first author on a paper recently published in Nature Communications. “The brain is so compact it can fit on top of your shoulders, whereas a modern supercomputer takes up a space the size of two or three tennis courts.”
Like synapses connecting the neurons in the brain that can be reconfigured, the artificial neural networks the team is building can be reconfigured by applying a brief electric field to a sheet of graphene, the one-atomic-thick layer of carbon atoms. In this work they show at least 16 possible memory states, as opposed to the two in most oxide-based memristors, or memory resistors [emphasis mine].
“What we have shown is that we can control a large number of memory states with precision using simple graphene field effect transistors [emphasis mine],” Das said.
The team thinks that ramping up this technology to a commercial scale is feasible. With many of the largest semiconductor companies actively pursuing neuromorphic computing, Das believes they will find this work of interest.
Given R. Stanley Williams’s presence on the author list, it’s a bit surprising that there’s no mention of memristors. If I read the signs rightly the interest is shifting, in some cases, from the memristor to a more comprehensive grouping of circuit elements referred to as ‘neuristors’ or, more likely, ‘nanocirucuit elements’ in the effort to achieve brainlike (neuromorphic) computing (engineering). (Williams was the leader of the HP Labs team that offered proof and more of the memristor’s existence, which I mentioned here in an April 5, 2010 posting. There are many, many postings on this topic here; try ‘memristors’ or ‘brainlike computing’ for your search terms.)
In the September  issue of the journal Nature, scientists from Texas A&M University, Hewlett Packard Labs and Stanford University have described a new nanodevice that acts almost identically to a brain cell. Furthermore, they have shown that these synthetic brain cells can be joined together to form intricate networks that can then solve problems in a brain-like manner.
“This is the first study where we have been able to emulate a neuron with just a single nanoscale device, which would otherwise need hundreds of transistors,” said Dr. R. Stanley Williams, senior author on the study and professor in the Department of Electrical and Computer Engineering. “We have also been able to successfully use networks of our artificial neurons to solve toy versions of a real-world problem that is computationally intense even for the most sophisticated digital technologies.”
In particular, the researchers have demonstrated proof of concept that their brain-inspired system can identify possible mutations in a virus, which is highly relevant for ensuring the efficacy of vaccines and medications for strains exhibiting genetic diversity.
Over the past decades, digital technologies have become smaller and faster largely because of the advancements in transistor technology. However, these critical circuit components are fast approaching their limit of how small they can be built, initiating a global effort to find a new type of technology that can supplement, if not replace, transistors.
In addition to this “scaling-down” problem, transistor-based digital technologies have other well-known challenges. For example, they struggle at finding optimal solutions when presented with large sets of data.
“Let’s take a familiar example of finding the shortest route from your office to your home. If you have to make a single stop, it’s a fairly easy problem to solve. But if for some reason you need to make 15 stops in between, you have 43 billion routes to choose from,” said Dr. Suhas Kumar, lead author on the study and researcher at Hewlett Packard Labs. “This is now an optimization problem, and current computers are rather inept at solving it.”
Kumar added that another arduous task for digital machines is pattern recognition, such as identifying a face as the same regardless of viewpoint or recognizing a familiar voice buried within a din of sounds.
But tasks that can send digital machines into a computational tizzy are ones at which the brain excels. In fact, brains are not just quick at recognition and optimization problems, but they also consume far less energy than digital systems. Hence, by mimicking how the brain solves these types of tasks, Williams said brain-inspired or neuromorphic systems could potentially overcome some of the computational hurdles faced by current digital technologies.
To build the fundamental building block of the brain or a neuron, the researchers assembled a synthetic nanoscale device consisting of layers of different inorganic materials, each with a unique function. However, they said the real magic happens in the thin layer made of the compound niobium dioxide.
When a small voltage is applied to this region, its temperature begins to increase. But when the temperature reaches a critical value, niobium dioxide undergoes a quick change in personality, turning from an insulator to a conductor. But as it begins to conduct electric currents, its temperature drops and niobium dioxide switches back to being an insulator.
These back-and-forth transitions enable the synthetic devices to generate a pulse of electrical current that closely resembles the profile of electrical spikes, or action potentials, produced by biological neurons. Further, by changing the voltage across their synthetic neurons, the researchers reproduced a rich range of neuronal behaviors observed in the brain, such as sustained, burst and chaotic firing of electrical spikes.
“Capturing the dynamical behavior of neurons is a key goal for brain-inspired computers,” said Kumar. “Altogether, we were able to recreate around 15 types of neuronal firing profiles, all using a single electrical component and at much lower energies compared to transistor-based circuits.”
To evaluate if their synthetic neurons [neuristor?] can solve real-world problems, the researchers first wired 24 such nanoscale devices together in a network inspired by the connections between the brain’s cortex and thalamus, a well-known neural pathway involved in pattern recognition. Next, they used this system to solve a toy version of the viral quasispecies reconstruction problem, where mutant variations of a virus are identified without a reference genome.
By means of data inputs, the researchers introduced the network to short gene fragments. Then, by programming the strength of connections between the artificial neurons within the network, they established basic rules about joining these genetic fragments. The jigsaw puzzle-like task for the network was to list mutations in the virus’ genome based on these short genetic segments.
The researchers found that within a few microseconds, their network of artificial neurons settled down in a state that was indicative of the genome for a mutant strain.
Williams and Kumar noted this result is proof of principle that their neuromorphic systems can quickly perform tasks in an energy-efficient way.
The researchers said the next steps in their research will be to expand the repertoire of the problems that their brain-like networks can solve by incorporating other firing patterns and some hallmark properties of the human brain like learning and memory. They also plan to address hardware challenges for implementing their technology on a commercial scale.
“Calculating the national debt or solving some large-scale simulation is not the type of task the human brain is good at and that’s why we have digital computers. Alternatively, we can leverage our knowledge of neuronal connections for solving problems that the brain is exceptionally good at,” said Williams. “We have demonstrated that depending on the type of problem, there are different and more efficient ways of doing computations other than the conventional methods using digital computers with transistors.”
If you look at the news release on EurekAlert, you’ll see this informative image is titled: NeuristerSchematic [sic],
(On the university website, the image is credited to Rachel Barton.) You can see one of the first mentions of a ‘neuristor’ here in an August 24, 2017 posting.
According to researchers at Helmholtz-Zentrum Dresden-Rossendorf and the rest of the international team collaborating on the work, it’s time to look more closely at plasticity in the neuronal membrane,.
From the abstract for their paper, Intrinsic plasticity of silicon nanowire neurotransistors for dynamic memory and learning functions by Eunhye Baek, Nikhil Ranjan Das, Carlo Vittorio Cannistraci, Taiuk Rim, Gilbert Santiago Cañón Bermúdez, Khrystyna Nych, Hyeonsu Cho, Kihyun Kim, Chang-Ki Baek, Denys Makarov, Ronald Tetzlaff, Leon Chua, Larysa Baraban & Gianaurelio Cuniberti. Nature Electronics volume 3, pages 398–408 (2020) DOI: https://doi.org/10.1038/s41928-020-0412-1 Published online: 25 May 2020 Issue Date: July 2020
Neuromorphic architectures merge learning and memory functions within a single unit cell and in a neuron-like fashion. Research in the field has been mainly focused on the plasticity of artificial synapses. However, the intrinsic plasticity of the neuronal membrane is also important in the implementation of neuromorphic information processing. Here we report a neurotransistor made from a silicon nanowire transistor coated by an ion-doped sol–gel silicate film that can emulate the intrinsic plasticity of the neuronal membrane.
Especially activities in the field of artificial intelligence, like teaching robots to walk or precise automatic image recognition, demand ever more powerful, yet at the same time more economical computer chips. While the optimization of conventional microelectronics is slowly reaching its physical limits, nature offers us a blueprint how information can be processed and stored quickly and efficiently: our own brain.
For the very first time, scientists at TU Dresden and the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) have now successfully imitated the functioning of brain neurons using semiconductor materials. They have published their research results in the journal Nature Electronics (“Intrinsic plasticity of silicon nanowire neurotransistors for dynamic memory and learning functions”).
Today, enhancing the performance of microelectronics is usually achieved by reducing component size, especially of the individual transistors on the silicon computer chips. “But that can’t go on indefinitely – we need new approaches”, Larysa Baraban asserts. The physicist, who has been working at HZDR since the beginning of the year, is one of the three primary authors of the international study, which involved a total of six institutes. One approach is based on the brain, combining data processing with data storage in an artificial neuron.
“Our group has extensive experience with biological and chemical electronic sensors,” Baraban continues. “So, we simulated the properties of neurons using the principles of biosensors and modified a classical field-effect transistor to create an artificial neurotransistor.” The advantage of such an architecture lies in the simultaneous storage and processing of information in a single component. In conventional transistor technology, they are separated, which slows processing time and hence ultimately also limits performance.
Silicon wafer + polymer = chip capable of learning
Modeling computers on the human brain is no new idea. Scientists made attempts to hook up nerve cells to electronics in Petri dishes decades ago. “But a wet computer chip that has to be fed all the time is of no use to anybody,” says Gianaurelio Cuniberti from TU Dresden. The Professor for Materials Science and Nanotechnology is one of the three brains behind the neurotransistor alongside Ronald Tetzlaff, Professor of Fundamentals of Electrical Engineering in Dresden, and Leon Chua [emphasis mine] from the University of California at Berkeley, who had already postulated similar components in the early 1970s.
Now, Cuniberti, Baraban and their team have been able to implement it: “We apply a viscous substance – called solgel – to a conventional silicon wafer with circuits. This polymer hardens and becomes a porous ceramic,” the materials science professor explains. “Ions move between the holes. They are heavier than electrons and slower to return to their position after excitation. This delay, called hysteresis, is what causes the storage effect.” As Cuniberti explains, this is a decisive factor in the functioning of the transistor. “The more an individual transistor is excited, the sooner it will open and let the current flow. This strengthens the connection. The system is learning.”
Cuniberti and his team are not focused on conventional issues, though. “Computers based on our chip would be less precise and tend to estimate mathematical computations rather than calculating them down to the last decimal,” the scientist explains. “But they would be more intelligent. For example, a robot with such processors would learn to walk or grasp; it would possess an optical system and learn to recognize connections. And all this without having to develop any software.” But these are not the only advantages of neuromorphic computers. Thanks to their plasticity, which is similar to that of the human brain, they can adapt to changing tasks during operation and, thus, solve problems for which they were not originally programmed.
I highlighted Dr. Leon Chua’s name as he was one of the first to conceptualize the notion of a memristor (memory resistor), which is what the press release seems to be referencing with the mention of artificial synapses. Dr. Chua very kindly answered a few questions for me about his work which I published in an April 13, 2010 posting (scroll down about 40% of the way).
Many computational properties are maximized when the dynamics of a network are at a “critical point”, a state where systems can quickly change their overall characteristics in fundamental ways, transitioning e.g. between order and chaos or stability and instability. Therefore, the critical state is widely assumed to be optimal for any computation in recurrent neural networks, which are used in many AI [artificial intelligence] applications.
Researchers from the HBP [Human Brain Project] partner Heidelberg University and the Max-Planck-Institute for Dynamics and Self-Organization challenged this assumption by testing the performance of a spiking recurrent neural network on a set of tasks with varying complexity at – and away from critical dynamics. They instantiated the network on a prototype of the analog neuromorphic BrainScaleS-2 system. BrainScaleS is a state-of-the-art brain-inspired computing system with synaptic plasticity implemented directly on the chip. It is one of two neuromorphic systems currently under development within the European Human Brain Project.
First, the researchers showed that the distance to criticality can be easily adjusted in the chip by changing the input strength, and then demonstrated a clear relation between criticality and task-performance. The assumption that criticality is beneficial for every task was not confirmed: whereas the information-theoretic measures all showed that network capacity was maximal at criticality, only the complex, memory intensive tasks profited from it, while simple tasks actually suffered. The study thus provides a more precise understanding of how the collective network state should be tuned to different task requirements for optimal performance.
Mechanistically, the optimal working point for each task can be set very easily under homeostatic plasticity by adapting the mean input strength. The theory behind this mechanism was developed very recently at the Max Planck Institute. “Putting it to work on neuromorphic hardware shows that these plasticity rules are very capable in tuning network dynamics to varying distances from criticality”, says senior author Viola Priesemann, group leader at MPIDS. Thereby tasks of varying complexity can be solved optimally within that space.
The finding may also explain why biological neural networks operate not necessarily at criticality, but in the dynamically rich vicinity of a critical point, where they can tune their computation properties to task requirements. Furthermore, it establishes neuromorphic hardware as a fast and scalable avenue to explore the impact of biological plasticity rules on neural computation and network dynamics.
“As a next step, we now study and characterize the impact of the spiking network’s working point on classifying artificial and real-world spoken words”, says first author Benjamin Cramer of Heidelberg University.
A July 1, 2020 news item on ScienceDaily announces work which researchers are hopeful will allow them exert more control over neuromorphic devices’ speed of response,
“Neuromorphic” refers to mimicking the behavior of brain neural cells. When one speaks of neuromorphic computers, they are talking about making computers think and process more like human brains-operating at high-speed with low energy consumption.
Despite a growing interest in polymer-based neuromorphic devices, researchers have yet to establish an effective method for controlling the response speed of devices. Researchers from Tohoku University and the University of Cambridge, however, have overcome this obstacle through mixing the polymers PSS-Na and PEDOT:PSS, discovering that adding an ion conducting polymer enhances neuromorphic device response time.
Polymers are materials composed of long molecular chains and play a fundamental aspect in modern life from the rubber in tires, to water bottles, to polystyrene. Mixing polymers together results in the creation of new materials with their own distinct physical properties.
Most studies on neuromorphic devices based on polymer focus exclusively on the application of PEDOT: PSS, a mixed conductor that transports both electrons and ions. PSS-Na, on the other hand, transports ions only. By blending these two polymers, the researchers could enhance the ion diffusivity in the active layer of the device. Their measurements confirmed an increase in device response time, achieving a 5-time shorting at maximum. The results also proved how closely related response time is to the diffusivity of ions in the active layer.
“Our study paves the way for a deeper understanding behind the science of conducting polymers.” explains co-author Shunsuke Yamamoto from the Department of Biomolecular Engineering at Tohoku University’s Graduate School of Engineering. “Moving forward, it may be possible to create artificial neural networks composed of multiple neuromorphic devices,” he adds.
This is the second neuromorphic computing chip story from MIT this summer in what has turned out to be a bumper crop of research announcements in this field. The first MIT synapse story was featured in a June 16, 2020 posting. Now, there’s a second and completely different team announcing results for their artificial brain synapse work in a June 19, 2020 news item on Nanowerk (Note: A link has been removed),
Teams around the world are building ever more sophisticated artificial intelligence systems of a type called neural networks, designed in some ways to mimic the wiring of the brain, for carrying out tasks such as computer vision and natural language processing.
Using state-of-the-art semiconductor circuits to simulate neural networks requires large amounts of memory and high power consumption. Now, an MIT [Massachusetts Institute of Technology] team has made strides toward an alternative system, which uses physical, analog devices that can much more efficiently mimic brain processes.
The findings are described in the journal Nature Communications (“Protonic solid-state electrochemical synapse for physical neural networks”), in a paper by MIT professors Bilge Yildiz, Ju Li, and Jesús del Alamo, and nine others at MIT and Brookhaven National Laboratory. The first author of the paper is Xiahui Yao, a former MIT postdoc now working on energy storage at GRU Energy Lab.
That description of the work is one pretty much every team working on developing memristive (neuromorphic) chips could use.
On other fronts, the team has produced a very attractive illustration accompanying this research (aside: Is it my imagination or has there been a serious investment in the colour pink and other pastels for science illustrations?),
A June 19, 2020 MIT news release, which originated the news item, provides more insight into this specific piece of research (hint: it’s about energy use and repeatability),
Neural networks attempt to simulate the way learning takes place in the brain, which is based on the gradual strengthening or weakening of the connections between neurons, known as synapses. The core component of this physical neural network is the resistive switch, whose electronic conductance can be controlled electrically. This control, or modulation, emulates the strengthening and weakening of synapses in the brain.
In neural networks using conventional silicon microchip technology, the simulation of these synapses is a very energy-intensive process. To improve efficiency and enable more ambitious neural network goals, researchers in recent years have been exploring a number of physical devices that could more directly mimic the way synapses gradually strengthen and weaken during learning and forgetting.
Most candidate analog resistive devices so far for such simulated synapses have either been very inefficient, in terms of energy use, or performed inconsistently from one device to another or one cycle to the next. The new system, the researchers say, overcomes both of these challenges. “We’re addressing not only the energy challenge, but also the repeatability-related challenge that is pervasive in some of the existing concepts out there,” says Yildiz, who is a professor of nuclear science and engineering and of materials science and engineering.
“I think the bottleneck today for building [neural network] applications is energy efficiency. It just takes too much energy to train these systems, particularly for applications on the edge, like autonomous cars,” says del Alamo, who is the Donner Professor in the Department of Electrical Engineering and Computer Science. Many such demanding applications are simply not feasible with today’s technology, he adds.
The resistive switch in this work is an electrochemical device, which is made of tungsten trioxide (WO3) and works in a way similar to the charging and discharging of batteries. Ions, in this case protons, can migrate into or out of the crystalline lattice of the material, explains Yildiz, depending on the polarity and strength of an applied voltage. These changes remain in place until altered by a reverse applied voltage — just as the strengthening or weakening of synapses does.
The mechanism is similar to the doping of semiconductors,” says Li, who is also a professor of nuclear science and engineering and of materials science and engineering. In that process, the conductivity of silicon can be changed by many orders of magnitude by introducing foreign ions into the silicon lattice. “Traditionally those ions were implanted at the factory,” he says, but with the new device, the ions are pumped in and out of the lattice in a dynamic, ongoing process. The researchers can control how much of the “dopant” ions go in or out by controlling the voltage, and “we’ve demonstrated a very good repeatability and energy efficiency,” he says.
Yildiz adds that this process is “very similar to how the synapses of the biological brain work. There, we’re not working with protons, but with other ions such as calcium, potassium, magnesium, etc., and by moving those ions you actually change the resistance of the synapses, and that is an element of learning.” The process taking place in the tungsten trioxide in their device is similar to the resistance modulation taking place in biological synapses, she says.
“What we have demonstrated here,” Yildiz says, “even though it’s not an optimized device, gets to the order of energy consumption per unit area per unit change in conductance that’s close to that in the brain.” Trying to accomplish the same task with conventional CMOS type semiconductors would take a million times more energy, she says.
The materials used in the demonstration of the new device were chosen for their compatibility with present semiconductor manufacturing systems, according to Li. But they include a polymer material that limits the device’s tolerance for heat, so the team is still searching for other variations of the device’s proton-conducting membrane and better ways of encapsulating its hydrogen source for long-term operations.
“There’s a lot of fundamental research to be done at the materials level for this device,” Yildiz says. Ongoing research will include “work on how to integrate these devices with existing CMOS transistors” adds del Alamo. “All that takes time,” he says, “and it presents tremendous opportunities for innovation, great opportunities for our students to launch their careers.”
Coincidentally or not a University of Massachusetts at Amherst team announced memristor voltage use comparable to human brain voltage use (see my June 15, 2020 posting), plus, there’s a team at Stanford University touting their low-energy biohybrid synapse in a XXX posting. (June 2020 has been a particularly busy month here for ‘artificial brain’ or ‘memristor’ stories.)
Getting back to this latest MIT research, here’s a link to and a citation for the paper,
Protonic solid-state electrochemical synapse for physical neural networks by Xiahui Yao, Konstantin Klyukin, Wenjie Lu, Murat Onen, Seungchan Ryu, Dongha Kim, Nicolas Emond, Iradwikanari Waluyo, Adrian Hunt, Jesús A. del Alamo, Ju Li & Bilge Yildiz. Nature Communications volume 11, Article number: 3134 (2020) DOI: https://doi.org/10.1038/s41467-020-16866-6 Published: 19 June 2020
In 2017, Stanford University researchers presented a new device that mimics the brain’s efficient and low-energy neural learning process [see my March 8, 2017 posting for more]. It was an artificial version of a synapse — the gap across which neurotransmitters travel to communicate between neurons — made from organic materials. In 2019, the researchers assembled nine of their artificial synapses together in an array, showing that they could be simultaneously programmed to mimic the parallel operation of the brain [see my Sept. 17, 2019 posting].
Now, in a paper published June 15  in Nature Materials, they have tested the first biohybrid version of their artificial synapse and demonstrated that it can communicate with living cells. Future technologies stemming from this device could function by responding directly to chemical signals from the brain. The research was conducted in collaboration with researchers at Istituto Italiano di Tecnologia (Italian Institute of Technology — IIT) in Italy and at Eindhoven University of Technology (Netherlands).
“This paper really highlights the unique strength of the materials that we use in being able to interact with living matter,” said Alberto Salleo, professor of materials science and engineering at Stanford and co-senior author of the paper. “The cells are happy sitting on the soft polymer. But the compatibility goes deeper: These materials work with the same molecules neurons use naturally.”
While other brain-integrated devices require an electrical signal to detect and process the brain’s messages, the communications between this device and living cells occur through electrochemistry — as though the material were just another neuron receiving messages from its neighbor.
The biohybrid artificial synapse consists of two soft polymer electrodes, separated by a trench filled with electrolyte solution – which plays the part of the synaptic cleft that separates communicating neurons in the brain. When living cells are placed on top of one electrode, neurotransmitters that those cells release can react with that electrode to produce ions. Those ions travel across the trench to the second electrode and modulate the conductive state of this electrode. Some of that change is preserved, simulating the learning process occurring in nature.
“In a biological synapse, essentially everything is controlled by chemical interactions at the synaptic junction. Whenever the cells communicate with one another, they’re using chemistry,” said Scott Keene, a graduate student at Stanford and co-lead author of the paper. “Being able to interact with the brain’s natural chemistry gives the device added utility.”
This process mimics the same kind of learning seen in biological synapses, which is highly efficient in terms of energy because computing and memory storage happen in one action. In more traditional computer systems, the data is processed first and then later moved to storage.
To test their device, the researchers used rat neuroendocrine cells that release the neurotransmitter dopamine. Before they ran their experiment, they were unsure how the dopamine would interact with their material – but they saw a permanent change in the state of their device upon the first reaction.
“We knew the reaction is irreversible, so it makes sense that it would cause a permanent change in the device’s conductive state,” said Keene. “But, it was hard to know whether we’d achieve the outcome we predicted on paper until we saw it happen in the lab. That was when we realized the potential this has for emulating the long-term learning process of a synapse.”
A first step
This biohybrid design is in such early stages that the main focus of the current research was simply to make it work.
“It’s a demonstration that this communication melding chemistry and electricity is possible,” said Salleo. “You could say it’s a first step toward a brain-machine interface, but it’s a tiny, tiny very first step.”
Now that the researchers have successfully tested their design, they are figuring out the best paths for future research, which could include work on brain-inspired computers, brain-machine interfaces, medical devices or new research tools for neuroscience. Already, they are working on how to make the device function better in more complex biological settings that contain different kinds of cells and neurotransmitters.
Here’s a link to and a citation for the paper,
A biohybrid synapse with neurotransmitter-mediated plasticity by Scott T. Keene, Claudia Lubrano, Setareh Kazemzadeh, Armantas Melianas, Yaakov Tuchman, Giuseppina Polino, Paola Scognamiglio, Lucio Cinà, Alberto Salleo, Yoeri van de Burgt & Francesca Santoro. Nature Materials (2020) DOI: https://doi.org/10.1038/s41563-020-0703-y Published: 15 June 2020
Artificial Neural Network (ANN) is a type of information processing system based on mimicking the principles of biological brains, and has been broadly applied in application domains such as pattern recognition, automatic control, signal processing, decision support system and artificial intelligence. Spiking Neural Network (SNN) is a type of biologically-inspired ANN that perform information processing based on discrete-time spikes. It is more biologically realistic than classic ANNs, and can potentially achieve much better performance-power ratio. Recently, researchers from Zhejiang University and Hangzhou Dianzi University in Hangzhou, China successfully developed the Darwin Neural Processing Unit (NPU), a neuromorphic hardware co-processor based on Spiking Neural Networks, fabricated by standard CMOS technology.
With the rapid development of the Internet-of-Things and intelligent hardware systems, a variety of intelligent devices are pervasive in today’s society, providing many services and convenience to people’s lives, but they also raise challenges of running complex intelligent algorithms on small devices. Sponsored by the college of Computer science of Zhejiang University, the research group led by Dr. De Ma from Hangzhou Dianzi university and Dr. Xiaolei Zhu from Zhejiang university has developed a co-processor named as Darwin.The Darwin NPU aims to provide hardware acceleration of intelligent algorithms, with target application domain of resource-constrained, low-power small embeddeddevices. It has been fabricated by 180nm standard CMOS process, supporting a maximum of 2048 neurons, more than 4 million synapses and 15 different possible synaptic delays. It is highly configurable, supporting reconfiguration of SNN topology and many parameters of neurons and synapses.Figure 1 shows photos of the die and the prototype development board, which supports input/output in the form of neural spike trains via USB port.
The successful development ofDarwin demonstrates the feasibility of real-time execution of Spiking Neural Networks in resource-constrained embedded systems. It supports flexible configuration of a multitude of parameters of the neural network, hence it can be used to implement different functionalities as configured by the user. Its potential applications include intelligent hardware systems, robotics, brain-computer interfaces, and others.Since it uses spikes for information processing and transmission,similar to biological neural networks, it may be suitable for analysis and processing of biological spiking neural signals, and building brain-computer interface systems by interfacing with animal or human brains. As a prototype application in Brain-Computer Interfaces, Figure 2 [not included here] describes an application example ofrecognizingthe user’s motor imagery intention via real-time decoding of EEG signals, i.e., whether he is thinking of left or right, and using it to control the movement direction of a basketball in the virtual environment. Different from conventional EEG signal analysis algorithms, the input and output to Darwin are both neural spikes: the input is spike trains that encode EEG signals; after processing by the neural network, the output neuron with the highest firing rate is chosen as the classification result.
The second generation of the Darwin Neural Processing Unit (Darwin NPU 2) as well as its corresponding toolchain and micro-operating system was released in Hangzhou recently. This research was led by Zhejiang University, with Hangzhou Dianzi University and Huawei Central Research Institute participating in the development and algorisms of the chip. The Darwin NPU 2 can be primarily applied to smart Internet of Things (IoT). It can support up to 150,000 neurons and has achieved the largest-scale neurons on a nationwide basis.
The Darwin NPU 2 is fabricated by standard 55nm CMOS technology. Every “neuromorphic” chip is made up of 576 kernels, each of which can support 256 neurons. It contains over 10 million synapses which can construct a powerful brain-inspired computing system.
“A brain-inspired chip can work like the neurons inside a human brain and it is remarkably unique in image recognition, visual and audio comprehension and naturalistic language processing,” said MA De, an associate professor at the College of Computer Science and Technology on the research team.
“In comparison with traditional chips, brain-inspired chips are more adept at processing ambiguous data, say, perception tasks. Another prominent advantage is their low energy consumption. In the process of information transmission, only those neurons that receive and process spikes will be activated while other neurons will stay dormant. In this case, energy consumption can be extremely low,” said Dr. ZHU Xiaolei at the School of Microelectronics.
To cater to the demands for voice business, Huawei Central Research Institute designed an efficient spiking neural network algorithm in accordance with the defining feature of the Darwin NPU 2 architecture, thereby increasing computing speeds and improving recognition accuracy tremendously.
Scientists have developed a host of applications, including gesture recognition, image recognition, voice recognition and decoding of electroencephalogram (EEG) signals, on the Darwin NPU 2 and reduced energy consumption by at least two orders of magnitude.
In comparison with the first generation of the Darwin NPU which was developed in 2015, the Darwin NPU 2 has escalated the number of neurons by two orders of magnitude from 2048 neurons and augmented the flexibility and plasticity of the chip configuration, thus expanding the potential for applications appreciably. The improvement in the brain-inspired chip will bring in its wake the revolution of computer technology and artificial intelligence. At present, the brain-inspired chip adopts a relatively simplified neuron model, but neurons in a real brain are far more sophisticated and many biological mechanisms have yet to be explored by neuroscientists and biologists. It is expected that in the not-too-distant future, a fascinating improvement on the Darwin NPU 2 will come over the horizon.
I haven’t been able to find a recent (i.e., post 2017) research paper featuring Darwin but there is another chip and research on that one was published in July 2019. First, the news.
The Tianjic chip
A July 31, 2019 article in the New York Times by Cade Metz describes the research and offers what seems to be a jaundiced perspective about the field of neuromorphic computing (Note: A link has been removed),
As corporate giants like Ford, G.M. and Waymo struggle to get their self-driving cars on the road, a team of researchers in China is rethinking autonomous transportation using a souped-up bicycle.
This bike can roll over a bump on its own, staying perfectly upright. When the man walking just behind it says “left,” it turns left, angling back in the direction it came.
It also has eyes: It can follow someone jogging several yards ahead, turning each time the person turns. And if it encounters an obstacle, it can swerve to the side, keeping its balance and continuing its pursuit.
… Chinese researchers who built the bike believe it demonstrates the future of computer hardware. It navigates the world with help from what is called a neuromorphic chip, modeled after the human brain.
Here’s a video, released by the researchers, demonstrating the chip’s abilities,
The short video did not show the limitations of the bicycle (which presumably tips over occasionally), and even the researchers who built the bike admitted in an email to The Times that the skills on display could be duplicated with existing computer hardware. But in handling all these skills with a neuromorphic processor, the project highlighted the wider effort to achieve new levels of artificial intelligence with novel kinds of chips.
This effort spans myriad start-up companies and academic labs, as well as big-name tech companies like Google, Intel and IBM. And as the Nature paper demonstrates, the movement is gaining significant momentum in China, a country with little experience designing its own computer processors, but which has invested heavily in the idea of an “A.I. chip.”
If you can get past what seems to be a patronizing attitude, there are some good explanations and cogent criticisms in the piece (Metz’s July 31, 2019 article, Note: Links have been removed),
… it faces significant limitations.
A neural network doesn’t really learn on the fly. Engineers train a neural network for a particular task before sending it out into the real world, and it can’t learn without enormous numbers of examples. OpenAI, a San Francisco artificial intelligence lab, recently built a system that could beat the world’s best players at a complex video game called Dota 2. But the system first spent months playing the game against itself, burning through millions of dollars in computing power.
Researchers aim to build systems that can learn skills in a manner similar to the way people do. And that could require new kinds of computer hardware. Dozens of companies and academic labs are now developing chips specifically for training and operating A.I. systems. The most ambitious projects are the neuromorphic processors, including the Tianjic chip under development at Tsinghua University in China.
Such chips are designed to imitate the network of neurons in the brain, not unlike a neural network but with even greater fidelity, at least in theory.
Neuromorphic chips typically include hundreds of thousands of faux neurons, and rather than just processing 1s and 0s, these neurons operate by trading tiny bursts of electrical signals, “firing” or “spiking” only when input signals reach critical thresholds, as biological neurons do.
Tiernan Ray’s August 3, 2019 article about the chip for ZDNet.com offers some thoughtful criticism with a side dish of snark (Note: Links have been removed),
Nature magazine’s cover story [July 31, 2019] is about a Chinese chip [Tianjic chip]that can run traditional deep learning code and also perform “neuromorophic” operations in the same circuitry. The work’s value seems obscured by a lot of hype about “artificial general intelligence” that has no real justification.
The term “artificial general intelligence,” or AGI, doesn’t actually refer to anything, at this point, it is merely a placeholder, a kind of Rorschach Test for people to fill the void with whatever notions they have of what it would mean for a machine to “think” like a person.
Despite that fact, or perhaps because of it, AGI is an ideal marketing term to attach to a lot of efforts in machine learning. Case in point, a research paper featured on the cover of this week’s Nature magazine about a new kind of computer chip developed by researchers at China’s Tsinghua University that could “accelerate the development of AGI,” they claim.
The chip is a strange hybrid of approaches, and is intriguing, but the work leaves unanswered many questions about how it’s made, and how it achieves what researchers claim of it. And some longtime chip observers doubt the impact will be as great as suggested.
“This paper is an example of the good work that China is doing in AI,” says Linley Gwennap, longtime chip-industry observer and principal analyst with chip analysis firm The Linley Group. “But this particular idea isn’t going to take over the world.”
The premise of the paper, “Towards artificial general intelligence with hybrid Tianjic chip architecture,” is that to achieve AGI, computer chips need to change. That’s an idea supported by fervent activity these days in the land of computer chips, with lots of new chip designs being proposed specifically for machine learning.
The Tsinghua authors specifically propose that the mainstream machine learning of today needs to be merged in the same chip with what’s called “neuromorphic computing.” Neuromorphic computing, first conceived by Caltech professor Carver Mead in the early ’80s, has been an obsession for firms including IBM for years, with little practical result.
[Missing details about the chip] … For example, the part is said to have “reconfigurable” circuits, but how the circuits are to be reconfigured is never specified. It could be so-called “field programmable gate array,” or FPGA, technology or something else. Code for the project is not provided by the authors as it often is for such research; the authors offer to provide the code “on reasonable request.”
More important is the fact the chip may have a hard time stacking up to a lot of competing chips out there, says analyst Gwennap. …
What the paper calls ANN and SNN are two very different means of solving similar problems, kind of like rotating (helicopter) and fixed wing (airplane) are for aviation,” says Gwennap. “Ultimately, I expect ANN [?] and SNN [spiking neural network] to serve different end applications, but I don’t see a need to combine them in a single chip; you just end up with a chip that is OK for two things but not great for anything.”
But you also end up generating a lot of buzz, and given the tension between the U.S. and China over all things tech, and especially A.I., the notion China is stealing a march on the U.S. in artificial general intelligence — whatever that may be — is a summer sizzler of a headline.
ANN could be either artificial neural network or something mentioned earlier in Ray’s article, a shortened version of CANN [continuous attractor neural network].
Shelly Fan’s August 7, 2019 article for the SingularityHub is almost as enthusiastic about the work as the podcasters for Nature magazine were (a little more about that later),
The study shows that China is readily nipping at the heels of Google, Facebook, NVIDIA, and other tech behemoths investing in developing new AI chip designs—hell, with billions in government investment it may have already had a head start. A sweeping AI plan from 2017 looks to catch up with the US on AI technology and application by 2020. By 2030, China’s aiming to be the global leader—and a champion for building general AI that matches humans in intellectual competence.
The country’s ambition is reflected in the team’s parting words.
“Our study is expected to stimulate AGI [artificial general intelligence] development by paving the way to more generalized hardware platforms,” said the authors, led by Dr. Luping Shi at Tsinghua University.
Using nanoscale fabrication, the team arranged 156 FCores, containing roughly 40,000 neurons and 10 million synapses, onto a chip less than a fifth of an inch in length and width. Initial tests showcased the chip’s versatility, in that it can run both SNNs and deep learning algorithms such as the popular convolutional neural network (CNNs) often used in machine vision.
Compared to IBM TrueNorth, the density of Tianjic’s cores increased by 20 percent, speeding up performance ten times and increasing bandwidth at least 100-fold, the team said. When pitted against GPUs, the current hardware darling of machine learning, the chip increased processing throughput up to 100 times, while using just a sliver (1/10,000) of energy.
Shelly Xuelai Fan is a neuroscientist-turned-science writer. She completed her PhD in neuroscience at the University of British Columbia, where she developed novel treatments for neurodegeneration. While studying biological brains, she became fascinated with AI and all things biotech. Following graduation, she moved to UCSF [University of California at San Francisco] to study blood-based factors that rejuvenate aged brains. She is the co-founder of Vantastic Media, a media venture that explores science stories through text and video, and runs the award-winning blog NeuroFantastic.com. Her first book, “Will AI Replace Us?” (Thames & Hudson) will be out April 2019.
Onto Nature. Here’s a link to and a citation for the paper,
Towards artificial general intelligence with hybrid Tianjic chip architecture by Jing Pei, Lei Deng, Sen Song, Mingguo Zhao, Youhui Zhang, Shuang Wu, Guanrui Wang, Zhe Zou, Zhenzhi Wu, Wei He, Feng Chen, Ning Deng, Si Wu, Yu Wang, Yujie Wu, Zheyu Yang, Cheng Ma, Guoqi Li, Wentao Han, Huanglong Li, Huaqiang Wu, Rong Zhao, Yuan Xie & Luping Shi. Nature volume 572, pages106–111(2019) DOI: https//doi.org/10.1038/s41586-019-1424-8 Published: 31 July 2019 Issue Date: 01 August 2019
This paper is behind a paywall.
The July 31, 2019 Nature podcast, which includes a segment about the Tianjic chip research from China, which is at the 9 mins. 13 secs. mark (AI hardware) or you can scroll down about 55% of the way to the transcript of the interview with Luke Fleet, the Nature editor who dealt with the paper.
The pundits put me in mind of my own reaction when I heard about phones that could take pictures. I didn’t see the point but, as it turned out, there was a perfectly good reason for combining what had been two separate activities into one device. It was no longer just a telephone and I had completely missed the point.
This too may be the case with the Tianjic chip. I think it’s too early to say whether or not it represents a new type of chip or if it’s a dead end.