By Lauren Lim
The human brain is a delicate lump of neural tissue that lives in our heads. This neural tissue, however, keeps numerous vital and intricate body systems running, allows us to interact with the outside world, and gives us the capacity for complex thoughts and emotions. Now, researchers are trying to approach computing in a similar way by using a combination of electrical and chemical signals like neurons do. Far superior to current computational technology, the brain is energy efficient, has rapid processing speed, and can potentially store unlimited information. Neuromorphic computing is based on enhancing computers by mimicking the brain, which could lead to developments in AI technology and computers that can connect and interact directly with the brain.
Caption: Semiconductor chip (Getty Images)
As the word “semiconductor” might suggest, the main materials used in semiconductor chips are pure metalloids, like silicon or germanium, which are poorer conductors of heat and electricity than metals (Segal, 2022). The properties of these elements allow developers to control the flow of energy through the material. Since their invention in the mid-20th century, semiconductor chips have become the basis of how modern electronics operate, but researchers are approaching the limits of advancement in semiconductor technology. Semiconductor chips are unable to meet the high demands of modern-day computing challenges, especially with the focus on AI development (Gumyusenge et al., 2021). Despite rapid progress in recent decades, AI is still leagues away from reaching the level of complexity and energy efficiency that animal brains have. In addition to the limits of semiconductor chips, due to the way conventional computers operate, information has to be transported between locations for processing and calculation, which limits the computer’s efficiency (Rand, 2021). This has led researchers to develop neuromorphic computer chips to mimic the structure of the brain.
The nervous system is formed of about 85 to 100 billion neurons that are in constant communication with one another (Seladi-Schulman, 2019). Neurons connect and transmit signals at sites called synapses. When a neuron fires, an electrical impulse travels along the neuron’s axon. This electrical signal causes neurotransmitters, or chemical messengers, to be released into the gap between one neuron’s axon and another neuron’s dendrites. When they bind to dendrites, neurotransmitters either increase or decrease the electrical potential of the receiving neuron firing. Unlike a traditional computer, which separates memory and calculation, each neuron individually processes and stores its electrical state, making it more energy-efficient and memory dense.
Computers work algorithmically and rigidly with binary. On the other hand, the nervous system is a non-sequential network, so it can operate parallelly, which allows it to operate more flexibly and efficiently (Lutkevich, 2020). A neuron’s stored electrical state is also on a continuous scale and can be manipulated as such. “When you have classical computer memory, it’s either a zero or a one. We make a memory that could be any value between zero and one. So you can tune it in an analog fashion,” said Alberto Salleo, a materials scientist at Stanford University (Kleiner, 2022). Of the neuromorphic technology in development, one type of device, memristive devices, focuses on storing its last electrical state. A simple polymeric memristor has two main layers. When exposed to an electrical current, one layer pushes positively charged ions into a second polymer layer, affecting the conductivity of that second layer. This mechanism allows the device to “remember” how much electricity has already been passed through the device, similar to neurons, even though memristors (Kleiner, 2022).
Salleo and other researchers have also been exploring the possibility of using organic materials to make devices able to substitute what were previously several components, such as memory cells and transistors (Kleiner, 2022). Other researchers are also developing similar devices, for example, a polymer device that can interact with neurons and change electrical state based on neurotransmitter output from the neurons. “Artificial neurons” like this open up the potential for prosthetic devices or computers that can read neural impulses, which is especially groundbreaking since current technology can only attempt this with imprecise and inconvenient electrodes (Kleiner, 2022).
Caption: Diagram of a neuron that is releasing dopamine and a device that can take input from the dopamine. The electrical resistance of the device is determined by the rate of dopamine release by the neuron. (Kleiner, 2022)
With the limits of semiconductor technology in view, plus the recent shortage of chips due to the COVID-19 pandemic, an increase in demand, and international tensions, researchers are making promising progress in neuromorphic computing. Although it likely won’t fully replace current technology, the mimicry of age-old biological mechanisms creates new paths to creating more efficient and more easily-applicable devices.
References
Britannica, T. Editors of Encyclopaedia (2015, November 22). memristor. Encyclopedia Britannica. https://www.britannica.com/technology/memristor
Gumyusenge, A., Melianas, A., Keene, S. T., & Salleo, A. (2021). Materials strategies for organic neuromorphic devices. Annual Review of Materials Research, 51(1), 47–71. https://doi.org/10.1146/annurev-matsci-080619-111402
Hopkinson, M. (2015, August 27). With silicon pushed to its limits, what will power the next electronics revolution? Phys.org. Retrieved October 31, 2022, from https://phys.org/news/2015-08-silicon-limits-power-electronics-revolution.html
Kleiner, K. (2022, August 29). Making computer chips act more like brain cells. Scientific American. Retrieved October 31, 2022, from https://www.scientificamerican.com/article/making-computer-chips-act-more-like-brain-cells/
Lutkevich, B. (2020, February 11). What is neuromorphic computing? SearchEnterpriseAI. Retrieved October 31, 2022, from https://www.techtarget.com/searchenterpriseai/definition/neuromorphic-computing
Rand, D. (2021, May 12). What's this neuromorphic computing you're talking about? HPE. Retrieved October 31, 2022, from https://www.hpe.com/us/en/insights/articles/whats-this-neuromorphic-computing-youre-talking-about-2105.html
Segal, T. (2022, September 23). What is a semiconductor and how is it used? Investopedia. Retrieved November 9, 2022, from https://www.investopedia.com/terms/s/semiconductor.asp
Seladi-Schulman, J. (2019, August 7). How many nerves are in the human body? . Healthline. Retrieved October 31, 2022, from https://www.healthline.com/health/how-many-nerves-are-in-the-human-body#function-of-nerves
Comments