What do you need to make a machine intelligent?
Fundamentally, two things: algorithms that can learn from and predict on provided data and a processor chip that can run the algorithms. We have plenty of algorithms crunching big data, thanks to massive strides taken by machine learning recently. But, what about the chip? Are the current microprocessors capable of handling brain-like computational tasks? While the chips haven’t changed much since the AI (artificial intelligence) boom, there is intense ongoing research focused on redesigning more efficient, lower energy consuming microprocessor chips. These microprocessors mimicking the human brain circuitry are called neuromorphic chips and the relevant branch of science is called neuromorphic computing/engineering.
To understand the developmental basis of these chips, one needs to get a quick picture of how neurons, the brain’s smallest functional units, talk to each other. Any single neuron at a given point of time is capable of transmitting information at high speeds to several neurons via neuron-neuron connections called synapses. The signals are both analogous and digital with their intensities varying in a controlled, consistent manner. The processor chips in the market right now operate in digital (discrete 0 or 1) fashion. Moreover, they cannot perform parallel computations like the brain.
In January 2018, two teams of researchers from Massachusetts Institute of Technology (MIT) and National Institute of Standards and Technology (NIST) individually developed materials that could lead to a new breed of processors that connect with each other and perform machine learning tasks using very low energy – just like neuronal networks in the brain. The team at MIT designed crystalline silicon-germanium chips that can transmit analog signals with minimal loss in signal strength. Moreover, researchers at NIST developed an ‘artificial synapse’ that could connect two processors and transmit information at lightning speeds. The artificial synapse, which is essentially manganese nanoclusters within a silicon barrier, uses ultra-low power and functions just like the biological synapse – higher the communication between the 2 processors, the stronger becomes their synapse. Apart from research institutes, the semiconductor company, Intel Corporation, recently unveiled a neuromorphic self-learning research chip called Loihi.
So, although there are research advances in neuromorphic computing, various technical challenges need to be overcome before neuromorphic chips enter mainstream technology. To begin with, the biological neural network is itself not very well understood, thus, making replication on a chip difficult. Secondly, most of the special materials for such chips require specific temperatures for efficient functioning. The atomic structures of the conducting materials change at specific temperatures, a key requirement for efficient signal transmission in neuromorphic chips. Current electronic devices have circuits that can effortlessly run at high temperatures (100 degrees or even above). To employ neuromorphic chips, the materials should be able to change their state (be it crystalline, amorphous, or metallic) at appropriate temperatures, which is currently not the case. However, on these lines, a research duo from École Polytechnique Fédérale de Lausanne (EPFL) recently generated a semiconductor material that can function at elevated temperatures and outperforms silicon. These are germanium coated vanadium dioxide films. This holds potential applications in neuromorphic computing. The third major issue that needs to be tackled is the creation of about a million synapses for complex tasks. Clearly, the neuromorphic computing market is still in its infancy. But, the great news is that funding is plenty and several international computing giants along with leading technical institutes are already working towards resolving the challenges. It will most likely be a matter of short waiting time before the neuromorphic-computing ushers in the next wave of AI.
References
- Brain Communicates in Analog and Digital Modes Simultaneously
- Modulation of intracortical synaptic potentials by presynaptic somatic membrane potential
- A Population-Level Approach to Temperature Robustness in Neuromorphic Systems
- Ultralow power artificial synapses using nanotextured magnetic Josephson junctions
- SiGe epitaxial memory for neuromorphic computing with reproducible high performance based on engineered dislocations
- Breakthrough Material for Aerospace and Neuromorphic Computing
________________________________________________________________________________________________________________________________
Author
Saikata Sengupta is currently pursuing her Ph.D. from Department of Neurology at Friedrich Schiller University, Germany. You can follow her on Linkedin or Twitter
Editors
Roopsha Sengupta is a freelance manuscript editor and is trying to break into a suitable scientific editing and writing role. She did her Ph.D. in the Institute of Molecular Pathology, Vienna and postdoctoral research at the University of Cambridge UK, specializing in the field of Epigenetics. Besides science and words, she enjoys spending time with children, doodling, and singing.
Rituparna Chakrabarti pursued her Ph.D. in Neuroscience from Georg-August University (Göttingen, Germany) and is currently a post-doctoral fellow at the Center for Biostructural Imaging of Neurodegeneration (BIN), Göttingen. For her, the interface of Science and art is THE PLACE to be! To unwind herself she plays mandolin and eagerly looks for a corner at a coffee house to slide herself in with a good read or company.
Illustrator
Cover Image is made by Smriti Srivastava. Apart from pipettes, colors are her buddies. She is a scientist by profession and an artist at heart. She likes illustrating science, sketching, landscape painting, Madhubani art form, traveling and photography.
Inset image- Pixabay
The contents of Club SciWri are the copyright of PhD Career Support Group for STEM PhDs {A US Non-Profit 501(c)3}, PhDCSG is an initiative of the alumni of the Indian Institute of Science, Bangalore. The primary aim of this group is to build a NETWORK among scientists, engineers and entrepreneurs).
This work by Club SciWri is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
