Summary
Reversible computing is a model of computation where the computational process, to some extent, is time-reversible. All modern computing is irreversible because logic gates in a circuit take two bits as input and produce one bit as output. A bit is lost by the gate as heat. In transforming two bits into one, the previous two bits cannot be reconstructed and as such the operation is irreversible. A reversible gate however takes two bits as input and computes two bits as output, so no bits are lost as heat. After the operation, the previous two bits can be reconstructed making the process reversible. If a computation does not lose energy in the form of heat, such computers can be orders of magnitude more energy efficient. This is important because Landauer’s limit, the minimum amount of energy needed for a computer to perform a computational step is called the “Landauer limit”, and the energy consumption of electronic-based computers is about thousand times larger than the Landauer limit. it has been estimated that modern computers will reach the Landauer bound around 2080.
Viability (2)
In 1961, Landauer described the Landauer embedding, a technique for transforming irreversible computations into equivalent reversible ones, but he thought his machines could not reversibly get rid of their undo trails. Over the decades, progress has been made proving reversible Turing machines by Lecerf, Bennett, Fredkin arriving with Toffoli in the 1980s inventing the Toffoli gate (also called the controlled-controlled-not gate), the most convenient universal reversible logic primitive. Reversible computing concept devices has been developed including gates, circuits, architectures, languages and algorithms including in a quantum computer, but no practical device has yet been built. The engineering challenges of totally isolating a reversible computer from the environment may be close to impossible without nanotechnological tools.
Drivers (2)
The demand for low power electronics is massive and driving R&D into all sorts of “Beyond CMOS” technologies. Reversible computing is low of the priority list because of the hard engineering required compared to other low-power computing approaches. The need for reversible computing really comes as we approach the Landauer bound, which assuming Koomey’s Law holds, that is the number of computations per joule of energy dissipated doubles about every 2.6 years, then we have another 60 years to 2080.
Novelty (5)
At the high level, the benefits of reversing operations or rolling back has interesting applications for error correction and software reliability. There are other ways of achieving the same benefits ex-ante such as formal verification but nothing ex-post. At the low-level, reversible computing is superior to all known other solutions on the energy consumption dimension. There are numerous pathways to improve energy efficiency by a few orders of magnitude from alternative materials, architectures and logic gate switching processes. But nothing has the potential to deliver the same multiple order of magnitude improvements and get anywhere close to the Landauer limit.
Diffusion (1)
Reversible computing requires something close to a paradigm shift in computer science. Irreversible computing is so fundamental to computing, few are aware there is even an alternative. The challenge is that the entire computing stack needs to be rebuilt from logic gates and transistors all the way up to programming language and algorithms. For example, standard programming languages contain many structures that are irreversible, such as conditionals, loops, jumps, and function calls. So developers would need to learn and switch to a new programming language such as the EEL (Energy- Efficient Language). The benefits could be so vast for HPC for example, that a new tech stack is cost-effective, but it will take a long time to build out all the necessary components.
Impact (5)
Reversible computing is the only technology we know of that allows humanity to flourish beyond the Landauer limit and past 2080. If we fail to surpass the theoretical limit of energy consumption of computation then we cap humanity’s progress in 2080. Reversible computing may never well be a necessary condition for the speculative beyond yottaFLOP processing demand such as human colonisation of space, society-wide Whole Brain Emulation, and close-to-reality-fidelity Virtual Reality.
Relevant companies