By: Hamid Salehi (AMD), Rahul Maganti (Wormhole Contributor), and Csongor Kiss (Wormhole Contributor)
Today, Wormhole, a leading blockchain interoperability platform, is collaborating with AMD to make FPGA hardware accelerators available to the Wormhole ecosystem. By increasing the accessibility of accelerators to blockchain developers, this collaboration aims to make a meaningful difference in accelerating trust and value transfer in permissionless environments. AMD will also lend its deep expertise in hardware acceleration to help deliver speed and scalability to multichain applications being built with Wormhole.
As the blockchain economy undergoes rapid expansion and distinct chains continue to emerge, there is an increasing imperative to develop secure, multichain bridges. A heterogeneous solution, where applications are not limited to a single blockchain, but instead can be built across multiple chains, can minimize the risk of developing on the 'wrong' platform. As state and data on-chain becomes more diverse and complex, abstracting away the idiosyncrasies of interacting with individual chains eases operational burdens for users and developers alike, driving the development of highly composable, multichain applications.
Wormhole launched in August of 2021 and was one of the first interoperability platforms to connect a diverse set of blockchains, allowing end-users and applications to move data and assets freely across the Web3 ecosystem. Since then, Wormhole has served as a foundational tool for cost-efficient, multichain communications, connecting more than 30 distinct blockchain networks across more than 6 different runtimes. Today, Wormhole is one of the most popular interoperability solutions in the crypto ecosystem, having processed more than $30 billion USD in volume and close to 1 billion messages sent since its launch. It is the de facto choice for blue-chip crypto-applications, such as among others.
Wormhole Today
Today, Wormhole's interoperability platform is maintained by a distributed set of validators (called Guardians) in a Byzantine fault tolerant (BFT) way.
A message is produced when at least two-thirds of the Guardians observe the same event on a particular blockchain, thus, users and applications must not only trust the consensus mechanism of the originating blockchain but also the consensus mechanism of Wormhole itself.
Wormhole Looking Forward
Wormhole is embracing zero-knowledge (ZK) proofs as a mechanism to improve the trust assumptions of the protocol. However, one of the main challenges with ZKP-enabled protocols is that they significantly increase the compute workloads required to maintain the network.
So, while embracing zero-knowledge cryptography promises to significantly improve the trust assumptions of the Wormhole protocol, it also presents some non-trivial challenges. This is where AMD can help.
Accelerating ZK with AMD FPGAs
Today, AMD offers enterprise-grade FPGAs that are based on the most advanced architectures for scalable compute and flexibility for a breadth of applications. FPGAs can parallelize massive workloads, such as processing millions of multichain messages on Wormhole.
Figure 1: GPUs and FPGAs offer Massive Hardware Parallelism for Zero-Knowledge Cryptography
Zero-knowledge proof-system consists of several cryptographic primitives. Accelerating only a few of these primitives can produce limited improvements in the overall proof-generation pipeline. Amdahl’s Law states that the improvement in overall system performance that is achieved through the acceleration (e.g., parallelization) of a specific function or operation is constrained by the percentage of time the accelerated function is actively used. This is, in effect, a theoretical upper bound on the benefit of parallelizing a system, as shown in
Conceptual Representation of Amdahl’s Law
Source: AMD Internal Data
What does this mean for zero-knowledge acceleration? For most research and development on hardware acceleration for ZK applications, this technology has focused on two core primitives: Number Theoretic Transforms (NTTs) and Multi-scalar Multiplication (MSMs), which often dominate proof-generation latency. However, given a typical ZK STARK-proving pipeline that involves other cryptographic primitives, Amdahl's Law tells us that the possible theoretical speedup-for the system by accelerating just the NTT and MSM components is limited usually to around 2.5x-5x.
To get proof-generation latency down to an order of seconds—not minutes or hours— Wormhole will need to focus on accelerating other components of its proof systems.
Wormhole is exploring techniques to accelerate a number of important zero- cryptographic primitives, including low-degree extensions (LDE), NTTs, transpose operations, and inner multiplications. Improving these operations may significantly reduce proof generation times.
Additionally, Wormhole contributors working with AMD technology are focused on speeding up FRI (Fast Reed Solomon Interactive Oracle Proofs of Proximity), an efficient Polynomial Commitment Scheme (PCS) as well as improving hashing performance and Merkle tree construction directly.
Simplified Diagram of a STARK-proving Pipeline
effort aims to deliver low-latency, trustless transfers that can scale with demand, enabling a seamless multichain experience across the Web3 ecosystem.
Looking to the Future
Wormhole is excited to collaborate with AMD to help realize its vision of a trust-minimized interoperability platform for Web3.
AMD will also continue to serve as a technology provider for the which is dedicated to fostering the development and use of zero-knowledge cryptography1.
Zero-knowledge cryptography is a foundational technology that is key to adoption of blockchain systems and applications. AMD FPGAs can help bridge the gap from theoretical research and prototypes to practical implementations of fast and scalable proof systems in production
- NO PURCHASE NECESSARY. The ZPrize 2023 competition is Sponsored by Zprize, AMD is not a Sponsor. All questions concerning the competition should be directed to ZPrize. Void where prohibited.