Historical Background and Development of Co-Processors

In the traditional computing field, a coprocessor is a processing unit that offloads complex tasks from the CPU. This technology is very common in the computer industry; for example, Apple's M7 motion coprocessor, launched in 2013, significantly improved the motion detection sensitivity of smart devices. The widely known GPU is a coprocessor concept proposed by Nvidia in 2007, primarily responsible for tasks such as graphic rendering. The GPU accelerates applications running on the CPU by processing compute-intensive code, and this architecture is referred to as "heterogeneous" or "hybrid" computing.

The main role of the coprocessor is to undertake complex tasks with high performance requirements, allowing the CPU to focus on handling more flexible and variable work.

There are two major issues restricting application development on the Ethereum network:

1. High Gas fees limit the scope of on-chain application development. A standard transfer operation requires 21,000 Gas, which is already the baseline for Gas fees on the Ethereum network. Other operations, such as data storage, consume even more Gas, which severely hinders the large-scale adoption of applications and users.

2. Smart contracts can only access the data of the most recent 256 blocks. In the future, with the upgrade of Pectra and the implementation of the EIP-4444 proposal, full nodes will no longer store past block data. This data loss makes it difficult for data-driven innovative applications to emerge, affecting the development of data-intensive applications like Tiktok and Instagram on the blockchain.

These issues reveal that computing power and data availability are the main reasons limiting the large-scale adoption of new computing paradigms. The Ethereum blockchain itself was not designed to handle large-scale computing and data-intensive tasks. To accommodate these applications, the concept of co-processors needs to be introduced. The Ethereum chain itself acts as the CPU, while the co-processors are similar to GPUs, handling computing and data-intensive tasks.

With the development of zero-knowledge proof technology, most co-processor projects are based on zero-knowledge proof as a technical foundation to ensure the credibility of off-chain computations by the co-processor.

The application range of ZK co-processors is very extensive, covering almost all real decentralized application scenarios, including social networking, gaming, DeFi, on-chain data-based risk control systems, oracles, data storage, large language model training and inference, etc. Theoretically, all the functionalities that Web2 applications can achieve can also be implemented on the blockchain by ZK co-processors, while Ethereum serves as the final settlement layer to ensure the security of the applications.

The definition of ZK co-processors in the industry has not yet been fully unified. ZK-Query, ZK-Oracle, ZKM, etc. can all be considered co-processors, as they assist in querying complete on-chain data, off-chain trusted data, and off-chain computation results. From this perspective, Layer 2 is essentially a type of co-processor for Ethereum.

Co-processor Project Overview

Currently, well-known co-processor projects mainly focus on three application scenarios: on-chain data indexing, oracles, and ZKML. Among them, general-purpose ZK virtual machine projects like Delphinus focus on zkWASM, while Risc Zero is dedicated to the Risc-V architecture.

Co-processor Technology Architecture

Taking the general-purpose ZK co-processor as an example, we focus on analyzing the technical architectures of three projects: Risc Zero, Lagrange, and Succinct, in order to understand the similarities and differences in the technical and mechanism design of such general virtual machines, thereby assessing the future development trends of co-processors.

Risc Zero

Risc Zero's ZK co-processor is called Bonsai, which builds a set of blockchain-agnostic zero-knowledge proof components. Bonsai is based on the Risc-V instruction set architecture, offering strong versatility and supporting multiple programming languages such as Rust, C++, Solidity, and Go.

The main features of Bonsai include:

1. Universal zkVM, capable of running any virtual machine in a zero-knowledge/verifiable environment.
2. ZK proof generation system that can be directly integrated into smart contracts or blockchain.
3. General rollup, distributing the computation results verified on Bonsai to the chain.

The core components of Bonsai include:

1. Prover Network: Receives and verifies ZK code, generates ZK proof.
2. Request Pool: Stores proof requests initiated by users.
3. Rollup Engine: Collects proof results and packages them for upload to the Ethereum mainnet.
4. Image Hub: A visual developer platform for storing functions and applications.
5. State Store: Off-chain state storage.
6. Proving Marketplace: The ZK proof computing power market in the industry chain.

Lagrange

Lagrange aims to build co-processors and verifiable databases that contain historical data on the blockchain, supporting the development of trustless applications. Its main features include:

1. Verifiable Database: Indexing on-chain smart contracts storage, storing state in the database.
2. Computing based on the MapReduce principle: zkMR architecture that supports parallel execution.

Lagrange's database design involves three parts: contract storage data, EOA state data, and block data. It creates a block data structure that is friendly to SNARK proofs, with each leaf node being a block header.

The calculation of Lagrange's ZKMR virtual machine is divided into two steps:

1. Map: Distributed machines map data to generate key-value pairs.
2. Reduce: Distributed computers compute the proof separately and then merge.

ZKMR can combine proofs of small computations into a proof of the overall computation, effectively expanding the proof capacity for complex calculations.

Succinct

The goal of Succinct Network is to integrate programmable facts into various aspects of blockchain development. It supports a variety of codes, including Solidity and specialized zero-knowledge domain languages, which can be executed in off-chain co-processors.

The off-chain ZKVM of Succinct is called SP (Succinct Processor), supporting Rust and other LLVM languages. Its core features include:

1. Recursive proof technology based on STARKs.
2. Support for wrappers from SNARKs to STARKs.
3. Precompiled-centric zkVM architecture.

Co-processor Project Comparison

When comparing general-purpose ZK co-processors, we primarily consider the following aspects:

1. Data indexing/synchronization capability
2. The technical approaches adopted (SNARKs vs STARKs)
3. Does it support recursive proofs?
4. Efficiency of the proof system
5. Ecological Cooperation Situation
6. Financing Background

Currently, the technical paths of mainstream projects are tending to converge, with many adopting wrappers from STARKs to SNARKs, as well as recursive proof technologies. Given that the proof generation of ZK algorithms is the most costly and time-consuming aspect, various projects are building networks of provers and cloud computing markets.

In cases where the technical paths are similar, the breakthroughs of the project may rely more on the strength of the team and the ecological resources support from the VC behind it, in order to capture a larger market share.

The Difference Between Co-processors and Layer 2

Unlike user-oriented Layer 2, the co-processor is mainly aimed at application development. It can serve as an acceleration component or a modular component, applicable in the following scenarios:

1. As a chain off-chain virtual machine component of ZK Layer2
2. Provide off-chain computing power for applications on the public chain
3. Oracle for public chain applications to obtain verifiable data from other chains
4. Act as a cross-chain bridge for message transmission

The coprocessor brings the potential for real-time synchronization of data across the entire blockchain and high-performance, low-cost trusted computing, capable of reconstructing most middleware in blockchain, including oracles, data queries, cross-chain bridges, etc.

Challenges Faced by Co-Processors

1. The entry barrier for developers is high, requiring mastery of specific languages and tools.
2. The industry is in its early stages, and the performance of zkVM involves multiple complex dimensions.
3. The hardware and other infrastructure are not yet mature, and commercial implementation still requires time.
4. The technical paths of various projects are similar, making it difficult to form significant advantages, and the focus of competition has shifted towards resource and ecological collaboration.

Summary and Outlook

ZK technology has strong versatility and helps the Ethereum ecosystem develop from decentralization to trustlessness. ZK co-processors, as an important tool for the implementation of ZK technology, can theoretically realize a blockchain version of any Web2 application.

The large-scale adoption of ZK co-processors mainly depends on two factors: a fully chain real-time provable database and low-cost off-chain computation. Achieving this goal requires gradual iterative implementation. The commercial application of ZK computing power chips is a key prerequisite for the large-scale deployment of co-processors.

The current market cycle lacks innovation, providing an opportunity window for the construction of next-generation large-scale application technologies. It is expected that in the next cycle, the ZK industry chain is likely to achieve commercial implementation. Now is the best time to focus on core technologies that can support on-chain interactions for 1 billion users.