Skip to content

Instantly share code, notes, and snippets.

@zmanian
Created March 1, 2026 01:18
Show Gist options
  • Select an option

  • Save zmanian/adb4d41a725fe64ccfeb9a435e6d22bd to your computer and use it in GitHub Desktop.

Select an option

Save zmanian/adb4d41a725fe64ccfeb9a435e6d22bd to your computer and use it in GitHub Desktop.
Zcash PoW Algorithm Change Tradeoffs: Equihash, Blake3, and ZK-Friendly PoW Analysis

Zcash Proof of Work: Algorithm Change Tradeoffs and ZK-Friendly PoW Analysis

Research compiled February 2026

Table of Contents

  1. The Core Problem: Equihash ASIC Monopoly
  2. Alternative PoW Algorithms
  3. What Makes ASIC Markets Competitive
  4. Blake3 Deep Dive
  5. ZK-Friendly Proof of Work
  6. Zcash History with ASICs
  7. Strategic Options
  8. Assessment

The Core Problem: Equihash ASIC Monopoly

Zcash's Equihash (200,9) has an effectively single-vendor ASIC market. Bitmain's Antminer Z15 Pro (840 kSol/s) is the only competitive hardware. Innosilicon's A9++ (140 kSol/s) is obsolete. No other manufacturers (IceRiver, MicroBT, Canaan, Goldshell) have Equihash product lines.

This directly caused the September 2023 ViaBTC incident where one pool accumulated 53.8% of hashrate, forcing Coinbase to raise ZEC confirmations to 110 blocks (~2.5 hours).

Why Equihash Failed at ASIC Resistance

The original 2016 thesis was that 144 MB working memory would make ASICs uneconomical. In practice:

  • Bitmain cracked it by May 2018 (Z9 Mini) — just 2 years after launch
  • 144 MB is manageable with on-chip SRAM at modern process nodes
  • The memory access pattern is predictable enough for hardware optimization
  • Equihash fails the key requirement: memory must exceed what's economically feasible to put on-die

Memory-hard functions resist ASICs only when: (a) access patterns are genuinely pseudorandom, (b) required memory exceeds what's economical on-die, and (c) bandwidth requirements approach external memory interface limits. Equihash fails condition (b) for modern semiconductor processes.


Alternative PoW Algorithms

Comparison Matrix

Algorithm ASIC Market Mining Access 51% Attack Cost Transition Risk Notes
SHA-256 Best (3+ vendors: Bitmain, MicroBT, Canaan, Bitdeer) ASIC only Extremely dangerous — BTC miners could trivially attack ZEC Very high Wrong choice for ZEC
KHeavyHash (Kaspa) Good (Bitmain, IceRiver, others) GPU initially, then ASIC Medium Medium Requires sufficient market cap to attract vendors
RandomX (Monero) None (CPU only) Any laptop Low — CPU rental is cheap Medium Botnet mining problem
Autolykos (Ergo) None yet (GPU) 4-6 GB GPU Low-Medium Medium Dynamic growing memory, proven GPU-only
Blake3 (Alephium) Good (4 vendors) ASIC rapidly dominated Medium Medium Simple, ASIC-friendly
New Equihash params Unknown (Bitmain likely catches up in 12-18 months) GPU briefly Medium Low Parameter fork, short ASIC reprieve
ZK-friendly (Poseidon2, RPO) Novel — different ASIC profile GPU/FPGA initially TBD High Most aligned with Zcash's mission

SHA-256

The most competitive ASIC market: 3+ manufacturers (Bitmain ~60-70%, MicroBT ~20-25%, Canaan ~5-10%), plus Bitdeer entering. Competition drives ~20% efficiency gains per generation. But Zcash's network is orders of magnitude smaller than Bitcoin's. Switching to SHA-256 would invite merge-mining and allow Bitcoin miners to trivially rent enough hashrate for a 51% attack.

KHeavyHash (Kaspa)

Matrix multiplication sandwiched between two Keccak hashes, designed to be ASIC-friendly. Market: Bitmain, IceRiver, and others compete. More competitive than Equihash because it emerged recently with high profitability. But at Zcash's market cap, uncertain whether multiple vendors would invest.

RandomX (Monero)

CPU-optimized via random code execution on a virtual machine. No viable ASIC path. But CPU hashrate is cheap to rent (cloud computing), lowering 51% attack cost. Monero has a significant botnet mining problem.

Autolykos (Ergo)

Dynamic memory table that increases in size over time, meaning ASICs built today become less efficient as memory requirements grow. No viable ASIC has emerged. The dynamic memory scaling is the most promising long-term ASIC deterrence mechanism, but Ergo is small-cap and unproven at scale.


What Makes ASIC Markets Competitive

Three factors determine whether multiple manufacturers enter:

1. Algorithm Simplicity = Lower Barrier to Entry

SHA-256 maps to standard digital logic. Any semiconductor team can design one. The result is 3+ major manufacturers. Equihash requires custom SRAM interfaces and a generalized birthday solver — enough complexity to deter smaller players.

2. Market Size (The Catch-22)

SHA-256 has 3+ vendors because Bitcoin is a $1T+ asset justifying massive R&D. Kaspa attracted multiple vendors at multi-billion peak market cap. Zcash at ~$400-600M may not justify independent ASIC development by more than 1-2 companies regardless of algorithm choice.

3. Memory Hardness Design

Memory-hard functions resist ASICs only when: (a) access patterns are genuinely pseudorandom, (b) required memory exceeds what's economical on-die, and (c) bandwidth requirements approach external memory interface limits. Autolykos's dynamic growing memory is the most promising approach.


Blake3 Deep Dive

Design and the Zooko Connection

Blake3 was co-designed by Zooko Wilcox-O'Hearn (along with Jack O'Connor, Jean-Philippe Aumasson, Samuel Neves), partially funded by ECC. It's a Merkle-tree-structured hash built on Blake2s's compression function with rounds reduced from 10 to 7, using pure Add-Rotate-XOR (ARX) operations on 32-bit words. No memory requirements, no S-boxes, no multiplications.

Internal parameters:

  • Word size: 32 bits
  • State size: 512 bits (16 x 32-bit words)
  • Block size: 64 bytes
  • Chunk size: 1,024 bytes (1 KiB)
  • Default output: 32 bytes (256 bits), extendable (XOF mode)
  • Security: 128-bit collision resistance, 256-bit preimage resistance

Blake3's tree structure means an ASIC can instantiate many identical compression pipelines, each independently processing a different 1 KiB chunk. For mining (block headers < 1 KiB), it degenerates to sequential — no tree parallelism benefit.

ASIC Market: Already More Competitive Than Equihash

Blake3 has 4 active ASIC manufacturers after just ~2 years of mining (vs. Equihash's 2 manufacturers over 8 years):

Manufacturer Top Model Hashrate Efficiency
Bitmain AL1 Pro 16.6 TH/s 0.225 J/GH
IceRiver AL3 15 TH/s 0.233 J/GH
Goldshell AL Max 8.3 TH/s 0.404 J/GH
DragonBall A40 3.3 TH/s 0.485 J/GH

The timeline from Alephium's launch (Nov 2021) to first ASIC (April 2024) was ~30 months. GPU mining became unviable within 6-8 months of ASICs appearing. The ASIC advantage is enormous: ~8,000x raw hashrate per device vs a top GPU.

Why Blake3 Attracts More Manufacturers

Blake3 is trivially simple to implement in silicon — no DRAM, no memory controllers, just replicated ARX logic pipelines. The design barrier is far lower than Equihash (which requires custom SRAM arrays consuming >90% of die area). Lower barrier = more entrants.

Blake3 vs Equihash Technical Comparison

Property Equihash (200,9) Blake3
Memory per thread 144 MB minimum ~512 bytes (state only)
Memory bandwidth dependency Critical bottleneck None
Solution size 1,344 bytes 32 bytes
ASIC design complexity High (memory arrays, controller) Low (pure logic)
ASIC memory area fraction >90% of die area ~0%
GPU advantage duration ~2 years (2016-2018) ~2.5 years (2021-2024)
ASIC manufacturers 2 (Bitmain dominant) 4 (Bitmain, IceRiver, Goldshell, DragonBall)

The Decred Lesson

Decred switched from Blake-256 to Blake3 in August 2023, simultaneously cutting PoW rewards to 1%. Their intent was to make ASIC development uneconomical for DCR alone. But because Alephium ASICs also mine Blake3, Decred got ASIC mining anyway via hardware built for a different chain. Shared algorithms create shared ASIC ecosystems.

Blake3 Bottom Line

Blake3 would give Zcash a more competitive ASIC market than Equihash (4 vendors vs 2). But it's explicitly ASIC-friendly with zero memory hardness. GPU miners would be eliminated quickly. Bitmain still leads. And it offers zero synergy with Zcash's ZK proof system.


ZK-Friendly Proof of Work

This is the most intellectually compelling direction for Zcash specifically, because it aligns mining incentives with the chain's core mission.

The Fundamental Inversion

Traditional hashes (SHA-256, Blake3, Equihash) use bitwise operations (XOR, rotation, addition mod 2^32) — trivially cheap in silicon, catastrophically expensive inside ZK circuits. ZK-friendly hashes (Poseidon, RPO, Monolith) use field arithmetic (multiplication mod large primes, power maps like x^5) — expensive in silicon, nearly free in ZK circuits.

What if Zcash's PoW function was the same math miners need for ZK proving? Mining investment would directly accelerate the ZK hardware ecosystem that Zcash's privacy depends on.

The Constraint Cost Gap

Hash R1CS Constraints Native Speed (vs SHA-256) Security Status
SHA-256 ~27,000 1x baseline Excellent (20+ years)
Blake3 ~20,000 2-4x faster Excellent
Poseidon2 ~240 ~5x slower Active bounty program
RPO ~280 ~10x slower Good, limited analysis
Griffin ~96 slower Partially broken (8/10 rounds)
Anemoi ~120 ~5x slower Partially broken (11/21 rounds)
Monolith lookup-based comparable to SHA-3 Very limited analysis

Poseidon requires ~100x fewer constraints than SHA-256 to prove in a ZK circuit. That's the synergy.

ZK-Friendly Hash Functions: Technical Survey

Poseidon / Poseidon2 — The dominant ZK-friendly hash. Uses HADES construction: partial rounds (one S-box) interspersed with full rounds (all S-boxes), plus MDS matrix mixing. x^5 S-box over BLS12-381 or BN254 scalar fields. Poseidon2 (2023) optimizes the linear layer for up to 70% fewer Plonk constraints and adds support for small fields (BabyBear, Mersenne31, Goldilocks). Most widely deployed (Filecoin, zkRollups, StarkWare, Penumbra). Active Ethereum Foundation cryptanalysis bounty program.

Rescue Prime Optimized (RPO) — Designed for STARK VMs like Polygon Miden. Operates over Goldilocks field (p = 2^64 - 2^32 + 1). Uses alternating power maps (x^alpha forward, x^(1/alpha) inverse) for better algebraic security than Poseidon. 12 field elements state, 7 rounds. Goldilocks arithmetic is hardware-friendly: reduction is just shifts and additions.

Monolith — Most aggressive native-performance ZK hash. Uses Keccak-derived chi S-boxes on sub-word bit arrays combined with prime fields. 7x faster than Poseidon2 for constant-time compression. First arithmetization-oriented function with plain performance comparable to SHA-3. But requires lookup arguments in proving systems, and has very limited cryptanalysis.

Griffin — Lowest R1CS constraints (96) but 8/10 rounds broken practically. Avoid.

Anemoi — Uses novel "Flystel" construction. 2x improvement over Poseidon in R1CS. But 11/21 rounds broken practically. Avoid.

Aleo Already Proved ZK PoW Works

Aleo launched mainnet in September 2024 with Proof of Succinct Work — miners generate Marlin zkSNARK proofs as their PoW puzzle. The dominant computations are MSM (multi-scalar multiplication) and NTT (number theoretic transform) over BLS12-377.

Result: Goldshell and IceRiver both ship ZK-specific ASICs optimized for field arithmetic. ZK-friendly PoW didn't prevent ASICs — it created ASICs whose optimization target (field multipliers, NTT units) overlaps with ZK proof acceleration hardware. This is the key empirical data point.

What ZK-Friendly ASICs Look Like

The silicon profile is fundamentally different from traditional PoW ASICs:

ASIC Type Gates per Core Op Design Complexity Barrier to Entry
SHA-256 (ARX) ~11-30K per round Low Low (competitive market)
Blake3 (ARX) ~2-5K per round (7 rounds) Low Low
Poseidon2 over BN254 ~629K per field multiplier High High
Poseidon2 over Goldilocks ~2-5K per multiplier Medium Medium

Goldilocks field (p = 2^64 - 2^32 + 1) is the sweet spot: reduction modulo p can be done with shifts and additions (no general division), making it 100-300x more area-efficient than BN254-scale arithmetic while still being ZK-native.

Three Visions for ZK-Friendly PoW on Zcash

A. Mining produces compact chain proofs (most practical)

Each block includes a ZK proof of chain state since genesis. Using IVC/folding schemes (Nova), miners produce incremental proofs as a byproduct of PoW. Enables instant sync: download one ~160-byte proof instead of validating the whole chain. Doesn't conflict with Zcash's privacy model.

B. Mining accelerates ZK hardware (Aleo model)

PoW difficulty drives investment into field-arithmetic ASICs. Those same ASICs make shielded transaction proving cheaper for users. Mining revenue subsidizes the ZK hardware ecosystem. Indirect but real benefit.

C. Mining IS shielded proving (most ambitious, most dangerous)

Miners generate Halo2 proofs for shielded transactions as PoW. Directly couples mining to privacy infrastructure. Major privacy conflict: miners see transaction witnesses unless witness-obfuscating outsourcing (WOO) is deployed. WOO adds overhead and only works for certain circuit types.

The Zcash Alignment Problem

Zcash's current stack uses Halo2 over Pasta curves (Pallas/Vesta) with Sinsemilla as the internal hash — not Poseidon. Aligning PoW with the proof system means either:

  1. Switch Zcash internals to Poseidon2 over Pasta — significant protocol change
  2. Use Goldilocks-based PoW — most hardware-friendly, but field mismatch with Pasta requires expensive conversion
  3. Accept the mismatch — mine with Poseidon2 over a convenient field, lose the "same hardware" synergy

Security: The Honest Risk Assessment

This is the critical concern for a $500M+ network:

Hash Years of Scrutiny Known Attacks Status
SHA-256 20+ None practical Gold standard
Blake3 6 (ChaCha lineage: 15+) None practical Very strong
Poseidon/2 5-7 Grobner basis better than claimed Active bounty, caution
RPO 4 Improved attacks reach 6/18 rounds Better, trending uncertain
Griffin 3 8/10 rounds broken Avoid
Anemoi 3 11/21 rounds broken Avoid
Monolith 2-3 Very limited analysis Too early for production

Every serious ZK-friendly hash has had significant cryptanalytic surprises within 1-3 years of publication. The Ethereum Foundation felt the need to fund a dedicated Poseidon bounty program — both reassuring (well-studied) and concerning (necessary).

Field Arithmetic Hardware Landscape

Companies building ZK-specific hardware acceleration:

  • Ingonyama ZPU: 72 Processing Elements at 1.305 GHz, each with modular multiplier. 13x area efficiency over A40 GPU for MSM.
  • Cysic C1: First-generation 12nm ASIC targeting MSM/NTT. ~1.31M Keccak proofs/second (13x GPU acceleration).
  • Supranational: Hardware acceleration for BLS12-381 operations.
  • Goldshell/IceRiver AE series: Production ZK ASICs for Aleo's PoSW.

The ZK hardware acceleration space is real and growing. A ZK-friendly PoW for Zcash would plug directly into this ecosystem.


Zcash History with ASICs

Timeline

  • 2016: Equihash chosen for ASIC resistance (144 MB working set assumption)
  • May 2018: Bitmain Z9 Mini announced — first Equihash ASIC. 80% community poll voted ASICs "bad for Zcash"
  • 2018-2019: Zcash Foundation governance vote — community voted against prioritizing ASIC resistance
  • 2019: Harmony Mining (dual-PoW) proposed for Blossom upgrade, never implemented
  • Sep 2023: ViaBTC accumulates 53.8% hashrate, Coinbase raises confirmations to 110 blocks
  • 2023-2024: Forum threads: "Bring Back GPU Miners?", "CPU/GPU Algo for Hybrid PoW/PoS"
  • 2024-2026: Crosslink/Trailing Finality Layer development (PoS overlay on PoW)
  • Jan 2026: ECC staff departure, governance restructuring

Key Lesson

The 2018 vote accepted that ASIC resistance was ultimately futile. The question isn't whether ASICs will exist — it's whether the ASIC market will be healthy and competitive.


Strategic Options

Option 1: Crosslink (PoS Finality Overlay) — Current Path

  • Keeps existing Equihash miners, no investment disruption
  • PoS finality makes 51% attacks insufficient even with concentrated hashrate
  • Reduces miner issuance ~50% over time
  • Risk: massive implementation complexity, new PoS attack surfaces
  • Status: Shielded Labs developing, milestone 4 in progress

Option 2: Switch to Blake3 (ASIC-Simple)

  • More competitive ASIC market (4 vendors vs 2)
  • Proven algorithm, excellent security
  • Risk: GPU elimination within months, no ZK alignment, Bitmain still leads
  • Zooko co-designed Blake3, but has not proposed it for Zcash PoW

Option 3: Switch to Autolykos-style (GPU-Friendly)

  • Maximizes miner decentralization via dynamic growing memory
  • Risk: hashrate collapse during transition, chain split, cheap GPU rental attacks

Option 4: ZK-Friendly PoW (Poseidon2/RPO)

  • Aligns mining with Zcash's core ZK mission
  • Mining investment accelerates ZK hardware ecosystem
  • Enables compact chain proofs and instant sync
  • Risk: cryptanalytic immaturity, protocol alignment complexity, transition cost
  • Novel — only Aleo has done this in production

Option 5: Hybrid (Memory-Hard + ZK Field Arithmetic)

  • Combine Equihash-style memory requirement with Poseidon2 field arithmetic
  • Slows ASIC convergence while maintaining ZK alignment
  • Most complex to design, least proven

Assessment

The Uncomfortable Truth About Market Cap

Even the best algorithm choice can't overcome a fundamental constraint: Zcash's ~$400-600M market cap probably can't sustain more than 1-2 ASIC vendors for any algorithm. SHA-256 has 3+ vendors because Bitcoin's hashrate market is worth billions annually. The ASIC market competitiveness problem is partly an algorithm problem and partly a market cap problem.

The Most Compelling Direction

ZK-friendly PoW is the most intellectually compelling option for Zcash specifically, because it aligns mining incentives with the chain's core mission (privacy via ZK proofs). No other chain has this natural alignment.

The most defensible path would be:

  1. Poseidon2 over Goldilocks as the PoW hash (best hardware efficiency, STARK-aligned, most cryptanalysis)
  2. Target chain state compression (Vision A) rather than shielded transaction proving
  3. Deploy with 2x the minimum security rounds
  4. Run a funded 1-2 year cryptanalysis competition before mainnet
  5. Consider a hybrid: memory-hard component + Poseidon2 field arithmetic to slow ASIC convergence
  6. Layer on top of Crosslink — PoS finality provides a safety net during the PoW transition

Why This Matters

Zcash choosing a ZK-friendly PoW would be the first time a major chain's mining function was deliberately designed to accelerate its own privacy infrastructure. Bitcoin miners secure Bitcoin but contribute nothing to Bitcoin's technical capabilities. Zcash miners could simultaneously secure the chain AND drive down the cost of shielded transactions by funding ZK hardware development through mining economics.

The question is whether the cryptanalytic maturity of ZK-friendly hashes is sufficient for a $500M+ network. The Ethereum Foundation's active Poseidon bounty program suggests the community isn't fully confident yet. A 1-2 year evaluation period with substantial bounties would be prudent before deployment.


This analysis was compiled from web research, academic papers, mining hardware databases, and community forum discussions. It reflects the state of knowledge as of February 2026.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment