Brevis presents Pico v1.0: the future of customizable and high-performance zkVMs for Web3 applications

1 month ago 20

Brevis, a platform for verifiable computing powered by zero-knowledge proofs, has just launched Pico, a modular and high-performance virtual machine (zkVM).

Pico allows the creation of specialized workflows for every need, integrating an advanced on-chain data coprocessor that improves performance up to 32 times.

Let’s delve deeper into the discussion below.

Brevis and the new challenges in the on-chain applications landscape

Brevis is a web3 entity that develops advanced technologies to optimize the use of Zero-Knowledge Proofs in the blockchain field.

Thanks to its advanced infrastructure, it allows cryptographic applications to transfer costly computational calculations to an off-chain environment, increasing scalability while preserving the security of the L1.

Up to today, several entities in the DeFi ecosystem (both chains and dApps) have successfully adopted the innovative features of Brevis, such as Kwenta, Usual, PancakeSwap, Frax, Quickswap, Mellow Finance, BeraBorrow, and others.

However, as Brevis engaged with new partners, new challenges and new needs in the context of customizable workflows emerged.

First and foremost, more and more projects have required unique priorities in terms of performance and complexity in business logic, with highly customizable functions.

Secondly, with the growth of new ZK technologies, it has been noticed that systems lacking a mod u lar architecture have often struggled with performance and optimizations.

Brevis has thus reached the conclusion of needing a zkVM virtual machine capable of adapting flexibly to the surrounding environment based on the complexity of each individual application, without having to use a rigid monolithic approach.

This is how Pico was born, a true turning point for computational calculation engines.

Brevis launches Pico: a revelation for web3 developers

Pico represents a zkVM with a modular architecture, designed to enable developers to build ZK applications tailored to their own needs.

Application developers of web3 can assemble Pico modules as if they were Lego bricks, thus adapting specifically to each computing requirement.

Whether it is about optimizing security, scalability, or efficiency in proof generation, this Brevis framework allows for creating specialized workflows for every need.

From a technical point of view. Pico features a “glue-and-coprocessor” architecture, based on advanced on-chain data processors capable of significantly enhancing performance. Additionally, it is compatible with the RISC-V standard and supports Rust programming toolchains, easily adapting to the toolsets used by developers.

By launching version v1.0 of Pico, Brevis introduced to the world the first zkVM that allows developers to test programs with:

Demonstration of the backends: STARK on KoalaBear and BabyBear and CircleSTARK on Mersenne 31.
Demonstration of workflows: optimize security, scalability, and proof generation objectives to meet the specific needs of your app.
Access to historical on-chain data: by integrating with the zkCoprocessor for on-chain data, developers are able to create dApps capable of accessing and processing historical on-chain data with complete security, offering the best performance and programmability.

Design and flexibility of the glue-and-coprocessor architecture

As mentioned, Pico is composed of a “glue-and-coprocessor” architecture that combines the efficiency of specialized co-processors with the flexibility of a general-purpose zkVM.

For those who do not know, co-processors are customized circuits that accelerate complex operations, such as arithmetic or cryptography. The ZkVM virtual machine acts as a glue in the process of proving and verifying data, ensuring that the logic not covered by the co-processors is still proven securely.

Furthermore, Pico uses precompilations (ready-made tools) that speed up some basic operations, such as hash or signature verifications. However, for more complex applications, precompilations alone are not sufficient.

For this reason, Pico has another level: it integrates a coprocessor for applications that handles specific operations related to data on blockchain, such as those used for Uniswap. This coprocessor further accelerates data access and verification, making ZK proofs faster, while maintaining great flexibility for developers who need to manage more complex logic.

To demonstrate the advantages of a similar hybrid approach, the Brevis team compared the performance of a “vanilla” Pico VM with a “boosted” Pico VM consecrated with the on-chain data zkCoprocessor.

The results showed a performance over 32 times higher for the Pico VM with consecrated coprocessor, with a cost equal to only 33% of the expenditure of the original Pico.

Despite the new version being even slower than the original, this zkVM represents an excellent compromise between performance and programmability, not available on the market today.

It is important to highlight how this general design model is not only applicable to use cases of on-chain computation and data access. In fact, Pico is integrating verifiable AI inference and Reth coprocessors at the app level to significantly accelerate the generation of proofs even for such cases.

Unprecedented Performance for Pico: the Comparison with Other zkVM Solutions

Compared to other existing zkVM solutions, Pico boasts more performant performance in terms of speed in proving operations on CPU.

The Brevis team compared Pico with the latest releases of RISC0, SP1, and OpenVM, thus comparing infrastructures with high workload.

The test was conducted on the same CPU machine, namely an AWS r7a.48xlarge instance with 192 CPU cores and 1.5 TB of RAM, allowing all zkVMs to generate proofs up to the final STARK proof before it is transformed into a SNARK.

Both the commonly used Fibonacci workload and two real-world scenarios were evaluated: Tendermint and a Reth block #17106222.

The results showed that Pico exhibits significant computational accelerations, with performance 155% better than the second-best solution.

These numbers establish a new speed benchmark for the zkVM world, opening new opportunities in the world of high-demand application development.

In all this, Brevis will soon release Pico’s GPU, so that it can handle high-intensity computational operations in parallel. Therefore, the performance results of the CPU will translate into calculations on the GPU.