Autonomous Matter
Hypersmart Matter
The Hypersmart Matter group creates (meta)materials and (meta)devices that process information in novel ways, using phenomena that are not traditionally used for computation. We aim to challenge the current approach to information processing, which assumes that every computational step is carried out perfectly. Instead, we want to build computers that embrace the complexity and imperfection of real-world physical systems.
Research focus
Our research is organized around four closely connected themes:
- We investigate how physical phenomena, such as self-sustaining oscillations and special wave states at the edges of materials, can enable entirely new ways of performing computations;
- We study the physical principles needed to build systems that can adapt, modify themselves, and learn from experience;
- We explore the fundamental limits of computation by investigating the minimum amount of energy required to process information;
- We develop intelligent physical devices that can directly process signals from the real world, without first converting them into a different form, making computation more efficient.
In conventional computers, information processing is primarily accomplished through the field effect, where a gate voltage modulates the conductivity of a silicon channel. There is significant potential in phenomena, such as noise, disorder, topology, plasticity, or symmetry, that are not traditionally used for computation. With my group, we seek to uncover how these phenomena can aid in the realization of more efficient computers.
Computers are physical objects and, as such, are limited by physical principles such as the laws of thermodynamics. However, experimentally probing these limits is challenging for two reasons: First, most physical systems are strongly dissipative (e.g., through ohmic losses); degrading their efficiency far below fundamental limits. Second, computing requires nonlinearities. Nonlinearities are only effective above a threshold energy, and for most experimentally realizable nonlinearities, this energy is much higher than thermodynamics dictates.
With my group, we are using (nano)optomechanical systems to explore the fundamental energy limits of computing. These systems can overcome both challenges: They present vanishing energy dissipation, and can display nonlinearities that are active at the level of the thermal noise—introduced either via geometric nonlinearity or optomechanically via the optical spring effect. Our approach is experimental and hands-on: We want to elucidate how practical limitations, for example in the kinds of nonlinear interactions that we can realize, determine the attainable computational performance.
Processing signals such as speech with digital computers requires first converting the information into the electronic domain. This conversion can be bulky (requiring transducers such as motors and strain gauges), inefficient (piezo microphones can have efficiencies that can dip below 1%) and energy-intensive (requiring batteries).
A key advantage of physical computers is that they can process information in its natural domain; for example, a neural network of mechanical resonators can in principle detect spoken commands without batteries or transducers. A soft robot can determine which direction to take without motors, processors, or batteries, through the nonlinear deformation of its body.
Processing real-world information, be it in a neural network or a digital circuit, requires a very large number of computing elements such as logic gates or neurons. Because of this, designing physical information processors presents unique challenges, and most works focus on small demonstrators and not complete tasks. With my group, we seek to develop the computational design tools necessary to design physical computers that can be scaled to real-world problems.
