EIP-2982 Serenity Phase 0: The Beacon Chain (Serenity Phase 0)

Final · Consensus & staking · EIP status Final · created 2020-09-15

Documents the first stage of the eth2 upgrade, in which a beacon chain running proof-of-stake (PoS) consensus is launched in parallel, while the existing proof-of-work chain is left untouched.

Why it is needed

Since the CryptoKitties surge in 2017, Ethereum blocks have remained consistently full and gas price markets have kept climbing. Simply raising the block gas limit increases bandwidth, computation, and disk burden, making it harder for ordinary users to run nodes and eroding decentralization. So eth2 planned a multi-stage path to scale by running multiple chains of similar capacity (shards) in parallel under a single consensus, while also switching that consensus itself to proof-of-stake. Proof-of-stake lowers the barriers to participation and technical requirements, increasing the decentralization of consensus, strengthening security through in-protocol penalties and economic finality, and eliminating energy-intensive mining. Phase 0 is the stage that safely bootstraps this new consensus without touching the current mainnet's consensus rules at all.

What changes

Today

After EIP-2982

Who is affected

In depth

Scope: a mainnet-unchanged bootstrap

Phase 0 is designed to require no breaking consensus changes to the existing Ethereum mainnet. It documents a path in which the new PoS consensus is bootstrapped in parallel and, once stabilized, replaces the current PoW consensus. The normative specification is not this EIP itself but lives in an independent repository, and the release tagged SPEC_RELEASE_VERSION v1.0.0, SPEC_RELEASE_COMMIT 579da6d2dc734b269dbf67aa1004b54bb9449784 is designated as the official specification for Phase 0. Phase 0's beacon chain is responsible only for maintaining its own consensus and managing the validator registry; it does not include ETH transfers, sharding, or smart contract execution. The plan describes later phases adding parallel shard chain consensus (Phase 1), merging in the existing Ethereum chain (Phase 1.5), and shard smart contract execution (Phase 2).

The deposit contract: the sole link between the two chains

Validator onboarding happens through the deposit contract DEPOSIT_CONTRACT_ADDRESS (0x00000000219ab540356cBB839Cbe05303d7705Fa) deployed on mainnet. The beacon chain reaches consensus on this contract's state to process new deposits, and this one-way deposit is the only technical link connecting mainnet and the beacon chain in Phase 0. A deposit includes a public key for signing messages, withdrawal credentials (a hash of the withdrawal public key), and the deposit amount, all signed together with the signing key. The signing key and withdrawal key are kept separate so that the higher-risk withdrawal key can be stored more safely offline and not shared with staking pools. The contract maintains a Merkle root of all deposits, and once that root is included in eth2 via Eth1Data voting, a block proposer submits a Merkle proof to begin the deposit process. MIN_GENESIS_TIME is 1606824000.

Consensus structure: Casper FFG and LMD GHOST

The beacon chain is built around two core components: the finality gadget Casper FFG (Casper the Friendly Finality Gadget) and the fork choice rule LMD GHOST (Latest Message Driven Greedy Heaviest Observed Sub-Tree). Casper FFG was chosen as the simplest algorithm available at the time, and being BFT-style, it carries a finality concept under which reverting a finalized block necessarily requires slashing at least one-third of validators. The details may change over the long term, and single-slot finality is also under exploration. LMD GHOST incorporates information from hundreds of validators every slot, making it extremely difficult to revert even a single block under normal conditions, and a block cannot be reverted unless an attacker controls close to 50% of the entire validator set. Proofs of safety and liveness are given in the Combining GHOST and Casper paper.

Issuance curve: proportional to the square root of total deposits

ETH issued as validator rewards is proportional to the square root of total stake, parameterized by BASE_REWARD_FACTOR 2**6 (64). This was chosen as a more stable and sustainable curve than the two alternatives, fixed total issuance and a fixed interest rate per ETH. The base reward takes the form k * D_i / sqrt(sum(D_j)): a fixed interest rate increases uncertainty in both total issuance and total stake, while fixed total reward is vulnerable to the discouragement attack, so this sits between the two. For example, at 1M ETH of active stake the maximum is 16.60% annually (166,208 ETH issued per year), and at 32M ETH it is 2.94% (940,167 ETH). These figures assume validators are 100% online and acting optimally, and this issuance is added on top of existing PoW rewards until the Merge.

Initial punitive parameters are deliberately lenient

The punitive parameters at launch were tuned to be less harsh than their final intended values. INACTIVITY_PENALTY_QUOTIENT is 2**26 (67,108,864), four times the final value, so the inactivity leak proceeds slowly during non-finality. The judgment is that non-finality in the first few months is far more likely to stem from client software issues than from a global catastrophe. PROPORTIONAL_SLASHING_MULTIPLIER is 1, one-third of the final value, which lowers the safety margin that can be held accountable during an attack, but early slashing is more likely to result from key management mistakes or client bugs than from a coordinated attack. MIN_SLASHING_PENALTY_QUOTIENT is 2**7 (128), four times the final value, keeping the minimum slashing penalty low. These settings are meant to encourage validator participation during the early, higher technical-risk stage.

The inactivity leak and anti-correlated slashing

If finalization stalls for more than 4 epochs (tsf > 4), a penalty is applied that drives the maximum reward to zero, and a second penalty proportional to tsf is added, so the total penalty grows quadratically over time. Even if more than 1/3 of validators go offline, offline validators' stake keeps shrinking until the online ratio recovers to 2/3 and the chain finalizes again. Under current parameters, if finalization halts, validators lose 1% of stake at 2.6 days, 10% at 8.4 days, and 50% at 21 days. Slashing has an anti-correlation structure: a detected validator loses stake equal to 3 times the proportion of validators slashed together around the same time (from 18 days before detection to around the time of withdrawal). This punishes coordinated attacks heavily while punishing isolated mistakes lightly, and also discourages everyone from pooling into the largest pool.

Why 32 ETH, why BLS

BFT consensus with accountable fault tolerance requires every validator to participate in two rounds before finalization, so for a validator count n and finalization time f, the per-second message overhead must be at least 2n/f. Taking total supply at roughly 2^27 ETH, with a 32 ETH stake unit there can be up to 2^22 validators, and with 2-epoch (768 second) finalization that comes to roughly 5461 messages per second at most. BLS signature aggregation is what makes this manageable: signatures on the same message can be combined through elliptic curve point addition alone, and the marginal cost of one more signature is 1 bit of data and one point addition. BLS is vulnerable to rogue key attacks, which are defended against by using the deposit message signature as a proof of possession. The epoch length of 32 slots is also a safety minimum that, even if randomness is manipulated, guarantees at least one honest proposer per epoch.

Lifecycle limits and Phase 0 withdrawal constraints

Activation occurs at least 4 epochs after the deposit is reflected (to prevent RANDAO manipulation), and the per-epoch cap on validators joining is max(4, |V|/65536), with any excess queued. Exits go through the same rate-limited queue. The entry and exit rate limits keep the validator set from changing too quickly between two points in time, so that as long as validators check in reasonably often (about once every 1 to 2 months if |V| >= 262144), finality guarantees are preserved. After leaving the exit queue there is also a roughly 27 hour withdrawal delay, during which violations can still be caught and slashed, and a slashed validator is delayed an additional roughly 36 days so that the number of validators slashed together over that period can be computed. However, in Phase 0 the 'withdrawn' state only exempts a validator from penalties and liability; there is actually nowhere to move the funds, and transfer functionality is provided in a later phase. This EIP is a Final-status Informational document; it does not change mainnet immediately, but it commits to this path being canonical.

Original EIP · ← EIPs