The most powerful, efficient computer known to man cannot be found in the depths of IBM, Google, Facebook, or any government agency — yet, everyone has access to it. How is that possible? Because the best computer in the world is the human brain. The brain consumes a mere 20 watts of power yet is capable of designing other computer networks, creating new languages, and understanding and manipulating matters it cannot see.
What is Neuromorphic Computing?
Our hardware-based computers and processors can't handle the same types of processing loads as the human brain. The goal of many programmers, hardware manufacturers, and data centers is to change that. The field of neuromorphic computing is dedicated to combining biology, electrical engineering, computer science, and mathematics technologies to create artificial neural systems capable of sensing and processing loads similar in capacity to the human brain and nervous system.
Neuromorphic Engineering: How it Began
The term 'neuromorphic engineering' was originally coined in the 1980s by Carver Mead, who has spent more than 40 years developing analysis systems aimed at mimicking the human body's senses and processing mechanisms, such as touching, seeing, hearing, and thinking. Neuromorphic computing is a subset of neuromorphic engineering that primarily focuses on the 'thinking' and 'processing' side of these human-like systems. While many people have never heard of neuromorphic computing technology, a more generalized technology that utilizes these systems and theories is widely known as artificial intelligence (AI).
What are the Goals of Artificial Intelligence?
While there are hundreds of interpretations, subsets, and theories that define what artificial intelligence actually is, the goal of all AI is to reproduce the functionalities of human behavior, thinking, and general tasks. Naturally, AI and neuromorphic computing are synonymous with each other in many ways as they each seek to replicate and even surpass human intelligence. In a sense, AI encompasses both neuromorphic computing and neuromorphic engineering, while embracing a variety of other technological facets. These include high-accuracy tasks such as detecting inconsistencies in manufacturing processes.
In the current state of technology, neuromorphic computing and AI are limited by the hardware capabilities on which these systems live. While Moore's Law continues to push these hardware technologies, they are nowhere near close to the capabilities of the human brain from a computational load perspective, let alone a power efficiency perspective. Though far from being human-like, these computational systems have taken vast strides in the last decade and will undoubtedly continue to bolster the goals of the neuromorphic computing world.
AI Hardware: Neuromorphic Computing Chips
A perfect neuromorphic chip that models the human brain identically might be perceived as a unicorn that will never be realized. However, as the computational scaffolding between humans and machines is fundamentally different (i.e., silicon is not grey matter), there are many lessons that the brain and its fundamental biology can teach computer scientists working on neuromorphic technology. Many chip architectures, both from a software and hardware perspective, have been heavily influenced by the findings of neuromorphic computing, which has led to many new versions of silicon architectures designed to achieve neuron-grade computing capabilities. More widely utilized chip technologies, such as FPGAs and ASICs, have been continually optimized by neuromorphic computing strategies and have even been used for AI workloads.
The Graphics Processing Unit (GPU) was initially designed to process graphics compute loads but was quickly utilized to develop AI algorithms over more widely adopted CPU technology, given their parallel load capability. GPUs have grown to be increasingly powerful and are often used as dedicated neural network accelerators to handle AI workloads. NVIDIA, widely regarded as a frontrunner in developing state-of-the-art GPU technology, has created dedicated edge computing devices, such as the NVIDIA Jetson Xavier Developer Kit, to run AI workloads for devices such as autonomous robots and facial recognition security systems.
Mathematics in AI software plays a significant role in executing AI programs. A mathematical object often utilized in these processes is called a tensor. Think of a tensor as a multidimensional matrix that can be scaled and changed as a function of the AI program. Without expanding into the theory behind AI program and algorithm structure, understand that tensor math is vital to modern artificial intelligence. It's so essential that Google created its own Tensor Processing Unit (TPU), which is a chip designed specifically for handling tensor math loads. While its TPU technology is utilized mostly in its data centers, Google has also created Google Coral products, which develop tensor-intensive programs.
Intel Labs, the research division of its famously CPU-centric company, has created the Loihi chip. This specially designed neuromorphic research chip aims to model and simulate a small-scale grey matter neural structure. The Loihi chip contains nearly 130,000 connected silicon 'neurons,' and it seeks to be the foundation of the next generation of AI hardware. In one instance, Intel Labs created a device called Pohoiki Beach, which combined 64 Loihi chips to create an 8 million neuron network. Developing a neural network program like that to run on this device is no small feat, and Intel Labs is continually optimizing programs for operation. In 2020, Intel plans on combining up to 100,000 Loihi chips to form a 100 million neuron neuromorphic computer, which, when ready, may well be the most powerful computer available.
The Future of Neuromorphic Computing
While still a budding subset of computer science, neuromorphic computing has yet to realize its full potential. Neuromorphic computing promises to be, at the very least, a powerful method of developing futuristic computing hardware and revolutionary AI software. If the technology proves to be the success that some claim it to be, neuromorphic computing may hold the secrets to consciousness and could be the last invention ever created by humans. This technology could paradoxically influence the research of the human brain, allowing for more accurate simulation and modeling of the soft grey matter between our ears that has engineered the world. It may create more intelligent, common-sense enabled algorithms that perform mundane tasks more efficiently than humans. Neuromorphic computing might just be the answer to self-driving cars and autonomous machines — only the future knows.