+6281216825037 [email protected]


hero honey

Let’s talk about neuromorphic computing. Whereas traditional computers have processors and memory as discrete components, neuromorphic computing attempts to emulate the brain in that each functional element—each synapse—is both processor and memory. This radically-different approach to computing requires new hardware for good performance, but it can provide incredible improvements in performance and efficiency for certain types of AI applications.

Neuromorphic computing isn’t a new concept by any means. IBM was producing neuromorphic chips back in 2014, and Intel came along with its Loihi chip in 2017, which received a follow-up last year. Samsung’s been experimenting with using its MRAM tech for neuromorphic computing, too. The point is, this stuff is still bleeding-edge research, but it’s not actually novel conceptually.
synapse
DARPA’s SynAPSE board uses 16 IBM TrueNorth neuromorphic processors.

What is novel is making memristors out of organic material. That’s exactly what some researchers from Washington State University have done. The specific medium that they have chosen is honey, owing to its relative stability (thanks to a lack of moisture, the same quality that makes it resistant to spoiling).

Using a thin sheet of honey between two metal electrodes, the researchers were able to create functional memristors. Moreover, the fabricated memristors demonstrated several properties that are useful for neuromorphic computing, including “spike-timing dependent plasticity” and “spike-rate dependent plasticity.”

lohi board
A shot of a board with numerous Intel Loihi neuromorphic processors.

Ultimately, however, these honey memristors are currently just a proof of concept. The ones created in the experiment exist on the micro-scale, while Intel and other companies are already fabricating synapses at the nano-scale. Even the densest neuromorphic processors, such as Intel’s Loihi 2, are still on the order of millions of neurons, while emulating the brain will take tens of billions of neurons.

The researchers say that they intend to pursue continued miniaturization, but emphasize that the primary benefit of their research is in allowing the creation of biodegradable, environmentally-friendly neuromorphic processors. Given that only the memristors themselves would be biodegradeable, it’s not completely clear what the practical value of such a benefit would be, but kudos to the scientists for thinking outside the box.



Source link