Ethereum Draft Proposes Unified Binary Tree, Showing Branch 768 Bytes Shorter
Key Facts
Draft EIP-8297 proposes moving Ethereum’s state tree to a partitioned binary structure, a change aimed at reducing state-proof sizes, according to the Ethereum Improvement Proposal created on June 11, 2026. The document describes itself as a draft in the Core Standards Track, so the structure is not an upgrade activated on the network. The proposal presents an illustrative comparison for a tree containing 2**24 elements: its binary branch is 768 bytes long, compared with 1,152 bytes for a quaternary branch. It links this difference to reducing the data needed to prove a path within the tree, targeting one aspect of the cost of verifying network state. A separate draft, EIP-8347, addresses the actual transition without setting an activation date.
The proposed structure uses a unified key-value space containing account headers, contract code and storage. The first byte of each key identifies the region to which the data belongs, according to EIP-8297. This brings the three categories together in one tree rather than treating them as separate structures. The draft says changing the tree does not alter the gas schedule and is invisible to the Ethereum Virtual Machine. It also says the change requires no modifications to contracts or Solidity and Yul code.
EIP-8347, created on July 23, 2026, sets out a proposed migration of state from the Merkle Patricia Trie to the proposed binary tree. The process begins with an offline conversion of state, after which nodes replay block-level access lists to synchronize the new tree with the chain. Nodes receiving a converted snapshot must rebuild the binary tree and verify its root, then rehash the leaves according to the old tree structure. The second check requires the state root for the reference block to be produced, allowing the converted state to be matched against the previous commitment. The proposal presents this as a way to verify the integrity of the conversion before adopting the new root.
The plan includes a monitoring period during which validators publish signed shadow roots for the binary tree for every block, while network consensus remains based on the Merkle Patricia Trie. During this period, nodes retain both trees and update them with each block; EIP-8347 says this roughly doubles state storage, as well as the work of committing state and read-write traffic. According to the draft, this is a transitional operating burden for nodes before the binary tree becomes the accepted commitment. The proposal also notes that removing each account’s storage root and changing the tree hash require coordination of updates to bridges, light-client verifiers and consumers of the eth_getProof interface. EIP-8347 names the activation fork but leaves its scheduling to a later proposal after the monitoring period ends.
This work is part of a broader discussion about expanding Ethereum’s capacity, but raising the gas limit and migrating the state tree are separate tracks in the published materials. The Ethereum Foundation said in its 2026 priorities update that the mainnet gas limit rose from 30 million to 60 million between two upgrades, and that the direction is to continue raising it toward 100 million and above, supported by block-level access lists and client testing. The roadmap lists the Glamsterdam upgrade in the fourth quarter of 2026, including access lists that map dependencies in advance to speed synchronization, execution and parallel reads from disk. The foundation describes the roadmap as a changeable plan reflecting researchers’ and developers’ best current estimate, and says node operators will need to update their clients in preparation for upgrades. State expiry and stateless-node support remain research topics that Ethereum.org expects to take years, with no guarantee that every proposal will be implemented.
A study published on October 27, 2026, offered an early empirical indication of how the tree could work with physical storage. A researcher on an Ethereum developers’ forum said they replayed two separate weeks of mainnet data in a partitioned binary tree and built two storage designs that produced the same state root as a reference implementation. The researcher added that six other designs were based on modeling the log and page flows generated by the replay and had not been tested on the database engines in question. These results keep the work within the scope of implementation and storage experiments, alongside the standards proposal and migration track. In the background to the discussion, Justin Drake presented Lean Ethereum on July 31, 2025, as his personal mission for the next decade, with long-term targets of 10K transactions per second for layer one and 10M for layer two. These are goals in the published vision, not a date or guarantee for activation of EIP-8297.