The Next Frontiers for Computing Chips

Over the past 50 years, advances in integrated circuit technology have fueled exponential progress in computing power, enabling ubiquitous technologies like smartphones, cloud services and artificial intelligence. However, continued progress under Moore’s Law is becoming more difficult as transistor sizes approach atomic scales. New innovations will be needed to extend the capabilities of computing chips.
One promising approach is to build chips with new materials that have better properties than silicon. Materials such as gallium nitride, silicon carbide and III-V semiconductors can operate at higher frequencies and handle higher power levels than silicon. This allows them to achieve faster switching speeds and higher integration densities. However, these exotic materials also pose manufacturing challenges and higher costs. They are currently being used to make high-performance RF chips and LEDs, while research continues on fabricating logic and memory chips with them.
Another option is to adopt new transistor designs. FinFET transistors have provided significant benefits in recent generations of chips, but they will reach scaling limits around 5 nanometers. Alternative 3D structures like nanowire, carbon nanotube and tunnel FET transistors could potentially replace the FinFET design. They provide a path to continue scaling to 3 nanometers and below. The main obstacles are optimizing performance and manufacturability as well as reducing costs. It may still take several years of research and development to make these new transistor types viable.
In addition, new chip architectures can unlock higher performance and efficiency gains. The ARM-based chips in smartphones and tablets are optimized for low power consumption, while GPUs feature massively parallel architectures targeted for graphics processing. New specialized architectures for artificial intelligence, encryption and in-memory computing are now emerging. More customized chip layouts and modular designs could also enable chips to better serve specific applications and workloads. However, designing and verifying these complex chips require state-of-the-art electronic design automation (EDA) tools.
Another exciting approach is neuromorphic computing, which uses chips designed like biological neural networks. Neuromorphic chips can process sensory data in an analog, event-driven fashion similar to neurons. This allows them to operate with much higher energy efficiency than conventional chips. IBM’s TrueNorth neuromorphic chip contains 5.4 billion transistors and consumes just 70 milliwatts of power. Qualcomm, Intel and various research groups are also developing neuromorphic chips for applications like computer vision, robotics, medical devices and wireless sensors.
In the long run, technologies like quantum computing could overcome the limits of transistor-based chips. A fully functional quantum computer could solve certain highly complex problems that are intractable for today’s supercomputers. But quantum computing is still in its infancy and requires major breakthroughs before achieving broad success and adoption. For now, continued progress in conventional computing chips remains crucial for powering technologies of the future. With innovation at multiple levels, from materials and transistors to chip architectures, computing is poised to keep advancing for decades to come.