Artificial neurons on silicon chips that behave just like the real thing have been invented by scientists – a first-of-its-kind achievement with enormous scope for medical devices to cure chronic diseases, such as heart failure, Alzheimer’s, and other diseases of neuronal degeneration.
Critically the artificial neurons not only behave just like biological neurons but only need one-billionth the power of a microprocessor, making them ideally suited for use in medical implants and other bio-electronic devices. The artificial neurons only need 140 nanoWatts of power.
They designed silicon chips that accurately modeled biological ion channels, before proving that their silicon neurons precisely mimicked real, living neurons responding to a range of stimulations.
Nature Communications – Optimal solid state neurons
Designing artificial neurons that respond to electrical signals from the nervous system like real neurons has been a major goal in medicine for decades, as it opens up the possibility of curing conditions where neurons are not working properly, have had their processes severed as in spinal cord injury, or have died. Artificial neurons could repair diseased bio-circuits by replicating their healthy function and responding adequately to biological feedback to restore bodily function.
In heart failure for example, neurons in the base of the brain do not respond properly to nervous system feedback, they in turn do not send the right signals to the heart, which then does not pump as hard as it should.
However developing artificial neurons has been an immense challenge because of the challenges of complex biology and hard-to-predict neuronal responses.
The researchers successfully modelled and derived equations to explain how neurons respond to electrical stimuli from other nerves. This is incredibly complicated as responses are ‘non-linear’ – in other words if a signal becomes twice as strong it shouldn’t necessarily elicit twice as big a reaction – it might be thrice bigger or something else.
Bioelectronic medicine is driving the need for neuromorphic microcircuits that integrate raw nervous stimuli and respond identically to biological neurons. However, designing such circuits remains a challenge. Here we estimate the parameters of highly nonlinear conductance models and derive the ab initio equations of intracellular currents and membrane voltages embodied in analog solid-state electronics. By configuring individual ion channels of solid-state neurons with parameters estimated from large-scale assimilation of electrophysiological recordings, we successfully transfer the complete dynamics of hippocampal and respiratory neurons in silico. The solid-state neurons are found to respond nearly identically to biological neurons under stimulation by a wide range of current injection protocols. The optimization of nonlinear models demonstrates a powerful method for programming analog electronic circuits. This approach offers a route for repairing diseased biocircuits and emulating their function with biomedical implants that can adapt to biofeedback.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
14 thoughts on “First as Artificial Neurons Developed for Medical Implants”
Interestingly, could these artificial neurons be used to treat symptoms of neuromotor diseases such as ALS, such as decreased respiratory capacity?
Neuron behavior is well understood. It is just like a transistor. If input reaches some threshold it fires. It can be modestly more complex. You can get a double strength action potential. It is basically trinary instead of digital. But double action potentials are not very common…and may even be more noise than message. There is no complex info in an action potential. It is just there or not. On-off. Though, as I said, you can have a very ON signal.
If you ignore the chemical signals, your artificial neuron won’t give the same response as a biological neuron would. And if you don’t produce chemical signals, your artificial neuron won’t evoke the same response from other biological neurons as a biological neuron would. May still be close enough for some applications, though.
I don’t see any issue with ignoring the inter-cellular biochemical, as long as you are not replacing large volumes of brain. And I think a lot of that is to insure the neurons are firing in a bounded range of frequency…mostly to protect the neurons. If they fire too infrequently they die (response to lack of need making room for connections for other neurons. this mostly happens early in life, perhaps also in depression), if they fire too frequently they die as well (more of a mechanical limit). This silicon stuff is probably much tougher, and won’t be self-destructing either way. There may be other roles for these chemicals, but these are not quick acting things.
82% of the brain in humans is cortex, so it should be on the lower end of your estimate or below.
I found a citation of between 0.3hz to 1.8hz average firing rate for the brain.
So maybe around 19W to 105W for a full brain emulation.
Oh – and the paper mentions that this design is “not optimized for low power consumption”…
I dug into the references a bit. The 140nW number (actually 139nW) is for neurons firing at 240hz – apparently a rate relevant to respiratory neurons.
And apparently nearly all of the energy was expended in the firing spikes, implying very low idle power per neuron.
If we use Mindbreaker’s cited 0.16 firings per second for cortical neurons, that’d be about 9.3W for continuous 100B neuron emulation.
I would guess some parts of the brain fire a lot more frequently, but still, it’s a lot closer to natural neuron efficiency than it might seem at first.
Not stated above, but impressive – the linked paper states that the energy used for these analogue electronic neurons is about 1 billion times lower than an equivalent digital implementation of a neuron.
Makes me think we may be on the verge of moving into a new wave of analog electronics.
Your calculation is for continuous firing of all neurons. You are ignoring that at any given moment, only a small fraction of neurons are doing anything. “…the average cortical neuron fires around 0.16 times per second.” https://aiimpacts.org/rate-of-neuron-firing/
Not really, no.
Roughly 100 billion neurons in a human brain that uses 20 watts. 140 nanowatts per artificial neuron.
14,000 watts for 100 billion neurons. Off by a factor of 700.
Neurons are like 10% of the brain…
Seems to me ….the real problem with modeling biological neurons in silicon is that biology can grow new circuits and connections… whereas silicon has a fixed capacity and you keep chewing up more and more capacity to model new connections until it Finally goes boom because there’s no new capacity left to form new connections in silicon… …It’s the same problem as FPGA’s…the more crap you put into an FPGA the less wiring resources it has left until you run out of wires then it has zero capacity left to grow …
Some observations from the abstract:
1. They seem to be focusing only on electrical signals, completely ignoring biochemical signals. They mention ion channels, but I’m not clear whether they actually sense local ion concentrations. I’m pretty sure they don’t have pumps to actively modify ion concentrations, let alone interact with other neurotransmitters. These are the same limitations as all current BCI candidate technologies, and will likely need to be addressed to fully emulate biological neurons.
2. Their focus is further on hippocampal and respiratory neurons. Specifically not cortical neurons. Do cortical neurons respond differently? What about different parts of the cortex (e.g. visual cortex vs frontal lobe)?
3. Their response functions are based on average behavior from a large sample of many biological neurons, and ignore any variability in individual neurons. We may need much better probing techniques to determine biological variability characteristics.
4. No mention of connectivity. Biological neurons can have several thousand connections each. How does this chip interface with the brain? Is it via a separate electrode grid, or some other method? (For that matter, do they even have an actual chip?)
5. Related to #4, no mention of resolution (if it needs a separate electrode, this would be determined by the electrode).
So, certainly a step forward, and may indeed be useful in some medical applications, but still a long way from any sort of artificial brain.
Um, that headline is a grabber. So is the rest.
Let’s hear it for the coming transhumanists, if it all holds up.
Comments are closed.