MIT Researchers Create 'Analog' Chip That Mimics Brain Synapses, Receptors

neuron synapse_Curtis Neveu

Readers may recall a post here on PS (August, 2011) concerning a new computer chip that ‘thinks’ like a brain cell (i.e., responds to new information),  due to its clever engineering and close linkage of processing and memory (see link to this article below). These were created through a joint IBM/DARPA project. These were referred to as neuro-synaptic chips.

Building a Better Neuron: From Digital to Analog:

As impressive as those chips were, they could hardly have been the ‘last word’ in neuromorphic computer chips, especially when you consider that they were not based on how brain synapses really work.

Now, from the  techno-genii at MIT, comes the latest advance in Artificial Intelligent (AI) computer chips: a chip that mimics a key feature of our adaptive brain cells: plasticity. Neural plasticity is fundamental to learning.

Notably, these newest chips mimic neuro-synaptic activity in ‘analog’ (not digital) form — much closer to the natural, analog tech of our brain cells — but are much faster than the biological version. The researchers believe that the chips will help them learn more about real neuronal activity, and, be used to control neural prosthetic devices like artificial retinas and limbs.

And, of course, as these chips come to mimic and surpass our own brain cells, true Artificial Intelligence* comes closer to reality.

Of Neurons, Synapses  & Ion Channels

Neurons, just like electrical circuits, get turned on and off; they get “excited” and “inhibited”. This activity is controlled by two main neurotransmitter chemicals: GABA (gamma amino butyric acid), which dampens signaling (inhibitory) between neurons,  and glutamate, which enhances signaling (excitatory) between neurons.

Synapses are the connections between neurons that permit the transfer of a signal from one to the other (a pre-synpatic neuron and a post-synaptic neuron), and using a complex system of neurotransmitter chemicals and ion channels.

Ion channels, embedded in the surface membrane at the synaptic gap between neurons, control the flow of positively charged atoms (cations) like sodium, potassium, calcium and fluoride.  As the channels open and close, they allow these ions to build up an electrical potential that may trigger a signal (an ‘action potential’) to be sent through the cell, and onto the next neuron (possibly).

These ion channels are key components underlying two brain processes known as long-term potentiation (LTP), and, long-term depression (LTD), which enable memory consolidation and learning.

 MIT neeural synapse chipBack to the Chip

The MIT chip contains about 400 transistors which allows it to mimic the behavior of a single brain synapse; each transistor is a true, analog version of an ion channel. The transistors allow a gradient of electrical potential to build up and drive the current through the chip, just as ions flow through a brain cell.

Though still embedded in silicon, the chip’s activity is a form of bio-mimicry, as opposed to a digital re-encoding/recreation.

“We can tweak the parameters of the circuit to match specific ion channels,” says Chi-Sang Poon, principal research scientist in the Harvard-MIT Division of Health Sciences and Technology.  “We now have a way to capture each and every ionic process that’s going on in a neuron.”

Previous attempts to build circuits that could simulate neural behavior, like its action potential, were somewhat successful, but they couldn’t simulate all the conditions that could trigger such a potential…they weren’t plastic enough.

 “If you really want to mimic brain function realistically, you have to do more than just spiking. You have to capture the intracellular processes that are ion channel-based,”  said Chi-Sang Poon in the MIT press release.

MIT researchers will use the chips to model neural functions like visual processing. Using a high-capacity computer, simulating even a single brain circuit can takes hours or even days to do digitally. But with the analog system, such simulations can be done even faster than real neurons process that (visual) information. They also foresee its use in a myriad of other AI applications.

LDP / LTP Puzzle Solved by Mimicking Marijuana-like Chemicals, Receptors:

How LTP and LDP work in neurons is a matter of long-standing debate; one theory states that both depend upon frequency of action potential (AP) in the post-synaptic cell), while another theory says that it’s all about the timing of  the AP’s arrival at the synapse.

What makes it difficult to determine what’s going on, and how each process is different,  is that both LTP and LDP require the same type of receptor – an NMDA receptor — to detect the post-synaptic AP. This doesn’t offer much help in distinguishing the two activities.

A new theory suggested that the two models could be unified if there were a second type of receptor at work in detecting the post-synaptic signal. The most promising candidate for this second receptor is the endo-cannabinoid receptor.

Endo-cannabinoids are naturally produced in the brain and are similar in structure to the cannabinoids in marijuana. They are involved in many functions, including appetite, pain sensation and memory. The theory is that chemicals are produced in the postsynaptic cell and then released into the synapse, where they activate presynaptic endo-cannabinoid receptors.

If NMDA receptors are active at the same time, then LTD occurs. According to the MIT press release:

When the researchers included on their chip transistors that model endo-cannabinoid receptors, they were able to accurately simulate both LTD and LTP.

Although previous experiments supported this theory, until now, “nobody had put all this together and demonstrated computationally that indeed this works, and this is how it works,” Poon says.

Chi-Sang Poon is the senior author of a paper describing the chip in the Proceedings of the National Academy of Sciences the week of Nov. 14. Guy Rachmuth, a former postdoc in Poon’s lab, is lead author of the paper. Other authors are Mark Bear, the Picower Professor of Neuroscience at MIT, and Harel Shouval of the University of Texas Medical School.

Some information for this post came from the MIT press release: ‘Mimicking the brain, in silicon‘ by Anne Trafton, MIT News Office

View the earlier Planetdsave article on the subject of ‘brain-like’ computer chips: New Computer Chip ‘Thinks’ Like a Brain Cell

Top image: Curtis Neveu ; CC – By- SA 3.0

Photo: (synapse/receptor mimicking computer chip) MIT


Leave a Reply

Your email address will not be published. Required fields are marked *