Ethereum 2.0 randomness
using a Verifiable Delay Function (VDF)
Core infrastructure
beacon chain
consensus layer
shards
application layer
secure sampling
EVM2.0 opcode
leader election
committee election
Key goals
unpredictable
unbiasable
unstoppable
Alternative randomness beacons
commit-reveal
RANDAO, Algorand, PoW
unpredictable
unbiasable
unstoppable
threshold crypto
Dfinity, Ouroboros (GOD)
unpredictable
unbiasable
unstoppable
RANDAO
slot 128
slot 2
slot 3
slot 4
beacon proposers
RANDAO epoch
128 slots (~17 minutes)
...
slot 1
...
global clock
RANDAO
slot 128
slot 2
slot 3
slot 4
RANDAO mix
beacon proposers
revealed?
RANDAO epoch
128 slots (~17 minutes)
...
slot 1
...
RANDAO
slot 128
slot 2
slot 3
slot 4
RANDAO mix
beacon proposers
⊕
revealed?
RANDAO epoch
128 slots (~17 minutes)
...
slot 1
...
RANDAO
slot 128
slot 2
slot 3
slot 4
RANDAO mix
beacon proposers
⊕
⊕
0x00
revealed?
RANDAO epoch
128 slots (~17 minutes)
...
slot 1
...
RANDAO
slot 128
slot 2
slot 3
slot 4
RANDAO mix
beacon proposers
⊕
⊕
⊕
0x00
revealed?
RANDAO epoch
128 slots (~17 minutes)
...
slot 1
...
RANDAO
slot 128
slot 2
slot 3
slot 4
RANDAO mix
beacon proposers
⊕
⊕
⊕
⊕
⊕
...
...
0x00
1-bit bias
revealed?
RANDAO epoch
128 slots (~17 minutes)
...
slot 1
or
0x00
...
Verifiable Delay Function (VDF)
VDF
input
32 bytes
output
256 bytes
Verifiable Delay Function (VDF)
VDF
difficulty
integer
input
32 bytes
output
256 bytes
Verifiable Delay Function (VDF)
VDF
difficulty
integer
input
32 bytes
output
256 bytes
proof
256 bytes�0.5ms verification
RANDAO + VDF
RANDAO mix
biasable
VDF output
unbiasable
1 epoch
RANDAO
mixing
VDF
delay
≥1 epoch
Safety argument
RANDAO
mixing
earliest
RANDAO mix
1 epoch
≥1 honest proposer
predictable by
attacker
Safety argument
RANDAO
mixing
VDF
delay
earliest
VDF output
earliest
RANDAO mix
1 epoch
≥1 honest proposer
Safety assumption
speed advantage ≤ Amax
vs
attacker hardware
commodity hardware
Liveness assumption
≥1 online VDF evaluator
Guaranteed delay
≥1 epoch
guaranteed delay
Amax epochs
evaluation period
Inclusion buffer
VDF
evaluation
RANDAO
mixing
1 epoch
1 epoch
Amax epochs
inclusion
buffer
onchain
offchain
RANDAO reseed
RANDAO
mixing
VDF
evaluation
RANDAO
mixing
inclusion
buffer
...
next 128 proposers
Parallel staggered beacons
Sequentiality assumption
x -> x2 % N
SQUARING
N # fixed 2048-bit RSA modulus
Sequentiality assumption
x -> x2 % N
SQUARING
N # fixed 2048-bit RSA modulus
x # VDF input
T # time parameter
x -> x2 -> x4 -> ... -> x2**T
T SQUARINGS
Wesolowski VDF (June 2018)
y = x2**T % N
OUTPUT
Wesolowski VDF (June 2018)
y = x2**T % N
p = x2**T//B % N
B # random 128-bit prime (Fiat-Shamir)
OUTPUT
PROOF
Wesolowski VDF (June 2018)
y = x2**T % N
p = x2**T//B % N
B # random 128-bit prime (Fiat-Shamir)
OUTPUT
PROOF
y == pB * x2**T%B % N
VERIFICATION
One-time RSA ceremony
1024
participants
trustless
coordinator
...
2048-bit
RSA modulus
N
Multi-party computation
Ligero Inc.
Participants | 1024 |
Modulus size | 2048 bits |
Security | (n – 1)-maliciously secure |
Synchronicity | synchronous |
Duration | ~10 minutes |
Communication | 20 rounds |
Rough expectations
Latency | 2 ns per modular squaring |
Process node | TSMC 16nm (FinFET) |
Die area | 20 mm2 |
Power | 7 W |
Form factor | standalone |
VDF ASIC count | 10 (= Amax, valid for 5 years) |
Cooling | fan |
Power | 100 W |
VDF ASIC
VDF rig
Costs
R&D | $15–$25 million |
Rigs | $5 million (5K rigs @ $1K each) |
Total | $20–$30 million |
in partnership with
Circuit competition
$1+ million
cash prizes
Open hardware research
VDF@ethereum.org
VDFresearch.org
PoW randomness vs VDF randomness
| Ethereum 1.0 (massively parallel) PoW | Ethereum 2.0 (inherently sequential) VDF |
Leader election | biasable | unbiasable |
Power consumption | ~2,300 MW | ~0.25 MW |
Hardware | ~10,000,000 GPUs | 2,500 VDF rigs |
Protocol subsidies | ~$1 billion per year | ~$1 million per year |
WW3-proof unbiasable randomness
largest multiparty computation
first open-source ASIC
largest cross-blockchain collaboration
thank you
and ~50 researchers and contributors
Question—What about exotic hardware?
→ not manufacturable for millions of gates
Question—Are any other projects interested in VDFs?