Massimult

Product Video

Computers can be built in many different ways. The core component of a computer is the CPU. All of the existing CPUs follow - more or less - certain architecture, called Von Neumann Architecture.

With Massimult, we use a completely radical approach that is based on combinator graph reduction. Program evaluation is intrinsically local and highly parallel. This enables quicker CPUs, which consume less energy, scale better, are more reliable, or difficult to attack.

In Massimult, you can imagine calculation as a tree in which all branches in parallel may change, driven by the leaves which are combinators. The tree grows and shrinks, and the result is ready when no further reduction is possible. For optimization, the tree becomes a graph, because identical subtrees are shared for efficiency.

In existing CPU architecture, programs and data are handled separately, putting data and codes into separate memory areas. This is significantly changed in Massimult as data and programs are handled uniformly. The core principle of current CPUs are arbitrary and globally that read and write. For better visualization, consider a toddler who puts things randomly from one place to another. Contrastingly, in Massimult everything is about local transformation. This local transformation has optimal properties regarding parallelity. So that every transformation in the graph that is possible at a certain time can happen in parallel with all other possible transformations without any conflict.

With the current architecture, a certain degree of parallelity is possible by the complicated construction of parallel programs by programmers. There is certain parallelity possible in the hardware design but it makes the chip construction extremely complicated.

Massimult is based on the theoretical model put forth by Alonso Church called Λ Calculus which was invented at almost the same time as Turing Machine. Λ Calculus has a twin sister that is based on the same principle as the former which is called Combinatory Logic. The idea to build the hardware on these principles is not new and was explored by Turner in the latter part of the 20th century. However, the problem of exponential growth of combinators was not solved at this time. With Massimult, we use a special family of combinators that has no exponential growth by are almost linear to Λ expressions.

There are certain benefits to this machine that are not offered in the existing solutions. For instance, Massimult is energy efficient and relies on a very less amount of energy. It is achieved through the locality of computation so that the fine-grained control of the activation of computing power is possible.