How it works

NovaNet is Modern and Adaptable

NovaNet is Modern and Adaptable

Historically zkVMs have been based on STARKish-like systems (Risc0, zkEVM). However, newer systems use completely different paradigms such as JOLT which can use any multivariate polynomial commitment scheme . NovaNet needs to work on a large variety of devices to achieve the end goal of permissionless decentralization. It also needs to work locally when privacy is the goal for application developers. These requirements lead to ICME's early work implementing non-uniform incremental verifiable computation schemes (NIVC). We started with the SuperNova proving scheme and modified it with opcodes and improvements such as optimal memory consistency checks (MCC), and an extension for parallelization using the same MCC paradigm.

Nova and similar IVC-based folding schemes are unmatched in memory efficiency. This means they take very little space to run. In NovaNet, the network decides on the size of the chunks a prover can handle based on computer specifications and the underlying execution trace being proved. We can effectively break a problem into smaller problems and run them in parallel. This allows specialized machines to work in some usecases, while smaller (potentially closer) network participants work on other portions of the proof. Cooperation, rather than proof racing, allows for NovaNet to handle a far wider range of usecases at amazing speeds. It also allows for better incentives for the network to adapt and grow. NovaNet system architecture is based on the very simple observation: winner-takes-all works for some use cases, while cooperation is better for many others.

NovaNet is a modular peer-to-peer decentralized prover network that any developer can leverage with ease. It is fast, cost effective, and highly adaptable.

Why peer-to-peer?

Some use cases require that proofs be run locally. For example, all applications that want complete privacy preservation need to first run on users' local machines. Any data given to a centralized provider leaks some information. Due to this fact, most centralized prover services that exist today cannot ensure full privacy. This greatly limits use cases.

Locality affects lag time in distributed networks. In the same local network, many machines may work together to solve a problem. With peer-to-peer this is made possible. In contrast, any centralized prover service, whether an L1 or L2, will need to move data across many hops in a network, regardless of problem set. They have no concept of the 'network edge'.

To truly say a network is permissionless anyone should be able to participate in it. NovaNet allows for this and further decentralization by allowing network participants to interact with one another directly. These traits allow for game theoretic optimizations that encourage cooperation, network growth, and ultimately faster, cheaper, and more flexible proving.