Researchers at the Massachusetts Institute of Technology have come up with a better way to simulate the processing that goes on in the human brain, and you hardware enthusiasts out there will appreciate this one.
Rather than simulate the firing and spiking of a bunch of neurons in software on massive clusters of computer chips, MIT researchers have created a digital chip with analog properties that emulates the flow of ions between connected brain cells and therefore can directly simulate how neurons actually fire across their synapses.
In a paper published this week in the Proceedings of the National Academy of Sciences, boffins describe a chip consisting of 400 transistors that mimics the ion flows in synapse between two real-world neurons and – they hope – will allow electronic circuits to mimic the “plasticity” that human brains have the ability to process, store, and adapt to new information. (Particularly when you concuss them or when their owners allow them to say something really stupid.)
Guy Rachmuth, a former postdoc Harvard-MIT Division of Health Sciences and Technology, is lead author of the paper, with Chi-Sang Poon, a principal research scientist at the lab, Mark Bear, a professor of neuroscience at MIT, and Harel Shouval of the University of Texas Medical School as co-authors.
The human brain has about 100 billion neurons, more or less (more for El Reg readers, and less for El Reg hacks), and each neuron has multiple synapses between them oozing neurotransmitters as the brain responds to stimuli from the outside world, creating an ion channel of flowing and charged sodium, potassium, and calcium ions in the synapse and eventually allowing an electric signal, called an action potential, to fire from one neuron to the other. When this happens, your brain remembers to do things, like duck a punch or keep your heart beating. (Often at the same time.)
MIT’s synapse emulation chip
What Rachmuth and his fellow boffins at MIT and UT have figured out how to do is to create a circuit that allows for current to flow through the transistors in an analog fashion, just like the ions in the ion channel in a synapse. And thus, the synapse chip emulate, in hardware, what a real synapse is doing in your head instead of relying on software to emulate all of this running on a cluster of ARM, Power, or x86 processors.
While other researchers have created chips that emulate the synaptic firing, this one can emulate the ion flows underlying the firing, and therefore do a better job simulating neurons. “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,” Poon explained in a statement announcing the paper. “We can tweak the parameters of the circuit to match specific ion channels. We now have a way to capture each and every ionic process that’s going on in a neuron.”
These ion channels are, explained Poon and Rachmuth in the paper, the key to two underlying pieces of brain microcode: long-term potentiation (LTP) and long-term depression (LTD). No, this is not what you had when you were a teenager and what you get as a grumpy old git when you don’t realize it.
Rather, these are the means by which changes in ion flows that strengthen or weaken the links between synapses. Those links are how we learn and forget things. By simulating the ion flows directly, the synapse chip made by MIT will be able to test theories about how LTP and LTD occur.
Some people think we learn based on the frequency of synaptic firing, others think the timing of sequences of firing are more important. Arrays of these synaptic chips will be able to show what works best empirically in hardware simulation and then extrapolate back to what happens in real brains.
Interestingly, there is a whole class of researchers who think that endo-cannabinoids, which have a structure similar to THC – the active ingredient in the pot you never inhaled – and which are involved in many brain functions including appetite, pain suppression, and memory, are affiliated with LTD. First, I knew my brain made its own opiates, but I did not know it made its own pot. And second, of course these endo-cannabinoids make you hungry and forget stuff – like the fact that you already knew that before you started designing the chip.
The MIT boffins are planning to use their synaptic chip to model specific parts of the brain, such as the visual cortex. And the upside is that compared to trying to simulate it in software on a supercomputer cluster, as researchers are now trying to do, by using the analog synaptic chip, the simulation will run faster than your own brain does. (The brain has a 65Hz to 80Hz cycle time, roughly, and it decreases with age, which is very likely why time seems to pass so slowly during grade school and so fast when you have finally got the kids out of the house.)
The interesting thing to ponder is how many synapses could be crammed onto a modern chip. The new “Interlagos” 16-core Opteron 6200 processors from Advanced Micro Devices have 2.4 billion transistors, so in theory, the 32 nanometer processes commonly used today could put at 6 million electronic synapses on the chips.
Forgetting the silicon you would need to interconnect the synaptic modules on the chip, or chip-to-chip interconnects and assuming 8,000 synapses per neuron, you would need around 133 million of these multi-synaptic chips to emulate the full brain and it would burn about 15.3 megawatts just for the chips alone, ignoring any other supporting electronics needed for the chips or the cooling for such a system. You brain does it in 20 watts, but it will think many orders of magnitude slower. Especially considering all the cannabis.
Source:http://www.theregister.co.uk/2011/11/17/mit_synaptic_chip/
