Computronium — A Fantasy about Computational Matter

However, at the end of the last century, when the prospects of nanotechnology seemed much more optimistic and achievable, futurists and engineers speculated about programmable substances in which each molecule serves as a computational unit. A machine like the T-1000 from the movie "Terminator 2: Judgment Day" could have been made from a hypothetical material of such a class; I tried to explore this topic on Habr in the article "Say a Word About Liquid Metal: Thoughts on the Hardware and Software Implementation of the T-1000." Such a robot made from mimicking polyalloy was a fantasy very much in the spirit of the time, as just five years before the film, Eric Drexler’s manifesto book (born 1955) "Engines of Creation: The Coming Era of Nanotechnology" was released (posted here). Drexler himself formulated the idea of molecular computing machines based on Richard Feynman's famous lecture "There's Plenty of Room at the Bottom: An Invitation to Enter a New World of Physics," delivered at Caltech on the eve of 1960. These proposals suggested that along with nanotechnology, even smaller "femto-technologies" could be realized. However, this viewpoint has not yet been confirmed, as at the atomic level, quantum effects dominate classical ones. Nevertheless, during its development, there was discussion about how realistic it would be to create "computronium" — a substance with maximum computational density, where each atom acts as a processor core.

Computronium and Cellular Automata

They proposed the term "computronium" ("computation") and described this hypothetical material in 1991 by Norman Margolus and Tommaso Toffoli from the Massachusetts Institute of Technology. They suggested that this material would be the first of its kind programmable matter, allowing the principles of computer modeling to create virtually any physical objects. Margolus's scientific interests included cellular automata, and the "machine" described in their paper with Toffoli was called "CAM-8," where CAM stands for "Cellular Automation Machine." According to the authors, a cubic meter of programmable matter could form a wind tunnel and then almost instantly restructure into a polymer mixture. It could be decomposed into a cloud of fermions, from which a specified pool of genes could be assembled, thus transitioning directly from physical experiments to epidemiological ones from the keyboard.

The idea originated as a thought experiment intended to popularize cellular automata but was soon picked up and began to evolve. In Margolus's view, pure computronium could represent a "computational crystal, which, by implementing the algorithms embedded in it, restructures itself according to the principles of a cellular automaton. As a result, the volumetric shape of the crystal changes, and all its molecules participate in computations." At the same time, Brossl Hasslacher from Los Alamos National Laboratory suggested that CAM-like substances could serve as raw materials for nanometer-scale supercomputers, where atoms of computronium would be arranged in uniformly distributed quantum dots and controlled from a terminal. When controlled according to the principles of cellular automata, such a system would exhibit self-organization and correct its own errors during modeling.

Over time, computronium came to be understood in a broader sense as the most efficient arrangement of a certain amount of matter, allowing the maximum use of the atoms of that matter for computations.

English physicist John Barrow, whose book I had the opportunity to translate into Russian, proposed the Barrow scale, which expands and specifies the famous Kardashev scale. The respected @SemenOk2 wrote about the Kardashev scale on Habr. This scale was proposed in 1967 by the Soviet physicist Nikolai Semyonovich Kardashev (1932 — 2019) to assess the level of development of extraterrestrial civilizations. Barrow suggested that advanced civilizations should not expand extensively, capturing more space and resources, but rather use the mastered space and resources more intensively. For example, around their native star, they could construct a “brain-matryoshka.” This hypothetical structure is an evolution of the concept of a "Dyson sphere," but it meets not the energy needs, but the computational needs of civilization. I previously made a brief overview of such hypothetical technologies in the publication "On Data Wars and Unfulfilled Noospheres." Ideally, the brain-matryoshka should consist precisely of computronium in its second interpretation — a substance maximally adapted for computations per unit mass. It is precisely about computronium in this interpretation that we will talk in the remaining part of the article.

Computational Power of Matter

So, let computronium be a substance whose structure allows it to approach the theoretical limit of computational power for a given volume of matter. That is, useful work in computations is performed by each atom of computronium. This material practically loses no energy in the form of heat, instead using it for computations.

Let's try to estimate how computational power can be calculated based on the mass of matter. In modern processors, it makes sense to measure it in FLOPs (floating-point operations per second). Another good measure is the clock frequency of the processor, which can indicate the pace of state changes in the system. The second metric in the case of computronium looks less appealing, since the material will need to spend some of those operations on calculations, and some on combating its own entropy, that is, on maintaining the "ingot" or "crystal" in a working state where it retains its ability to compute. On the other hand, there will be so many atoms in a macroscopic piece of computronium that the material will easily parallelize and/or pipeline any tasks, so the clock frequency from atom to atom may turn out to be uneven or variable.

On the other hand, modern computers (including the most advanced neural networks) were designed to perform mathematical operations that adhere to strict laws and can be executed iteration by iteration. It seems that computronium is significantly more inclined towards evolutionary algorithms and reinforcement learning, so its computational power may be expressed not just in the number of operations per second, but in the number of successful operations per second and in the number of iterations that yielded results, even if suboptimal or circuitous. Let’s not forget that computronium is most similar to a cellular automaton, so it likely should resonate in principle with the culture of xenobots, gray slime, or slime molds.

How to organize matter

of five to six types of atoms with the following functions:

1) Computational atoms

2) Atoms for storing information. This can refer to both permanent data storage and intermediate data used only during computations

3) Input/output atoms, network interface atoms, ensuring the interaction of computronium with regular computers or users - it's up to you whether this is one type of atom or two

4) Atoms-catalysts that initiate the necessary series of computations. They are likely functionally analogous to system calls.

5) Atoms that make up logical gates.

This substance would be more complex than a completely egalitarian slime mold, but as mentioned above, computronium should be compact while integrating with existing computing systems. In this case, the different types of computronium atoms are similar in specialization to various types of blood cells coexisting in plasma.

In this interpretation, computronium aligns with another hypothetical technology — “utility fog.” This concept, understood as a reconfigurable swarm of nanorobots, was proposed by John Hall in 1989, who believed that such a swarm could respond to external stimuli and instantly assemble, say, into “seat belts.” Utility fog may deserve a separate article on Habré, but here I will note its important distinction from computronium: the presence of sensory elements that respond to the state of the environment. That is, computronium with utility fog functions would not only perform computations but also switch from mode to mode as conditions change, with information about such changes obtained from yet another type of atom, supplementing those listed above.

How to make an atom binary

Thus, below the nanotechnology level, there is indeed room for further miniaturization of technologies, and computronium no longer seems as pure a science fiction as it did in 1991 (I remind you that Bose-Einstein condensate was first obtained in 1995, and the first quantum dot was constructed in 1993). Perhaps the only physical limitations for creating computronium that can be seen today are 1) the minimum temperature at which atomic movement continues (such as the temperature of cosmic microwave background, which is about 2.7 K) and 2) the Schwarzschild radius inherent to a quantum dot, beyond which it would turn into a miniature black hole. Hypotheses about whether information has mass, and whether a quantum dot could spontaneously develop an event horizon due to information overflow, definitely go beyond the scope of this article. However, when producing computronium, it would certainly be necessary to consider the Bekenstein bound — a quantity that allows calculating how much maximum information can be contained in a given region of space.

Taking all these inputs into account, I will classify computronium as second-class impossibilities according to Michio Kaku, and I look forward to seeing your opinions and ideas in the comments.