ENTROPY INFRASTRUCTURE — POST-QUANTUM READY

Certified entropy
at memory speed.

GPU-accelerated quantum entropy distillation with persistent streaming delivery for post-quantum cryptographic systems. No network round trip. No latency floor.

3.58 µs app latency
280× faster than REST
2.71→7.93 bits/byte H_min
NIST SP 800-90B inline
View Research Get in Touch
The Problem

Post-quantum migration exposes a structural entropy gap

NIST's 2024 finalization of FIPS 203 (ML-KEM) and FIPS 204 (ML-DSA) mandates post-quantum cryptography across federal systems. ML-KEM-768 requires 96 bytes of certified entropy per session — three times the 32 bytes needed by ECDH.

At 100,000 sessions per second, a realistic load for a large financial institution, entropy demand jumps from 76 Mbps to 153 Mbps overnight. Existing entropy services cannot support this migration.

REST API services (Qrypt, QuintessenceLabs) Every request waits for a network round trip. Minimum 1 ms per call. At 100,000 ops/sec the time budget per call is 10 µs. REST is structurally incompatible with this regime.
Hardware cards (ID Quantique Quantis PCIe) Certified and local but require in-house expertise, don't scale across sites, provide no managed health monitoring, and impose the full operational burden on the customer.
Denied environment deployments Tactical communications and autonomous systems operate where no network exists. All current entropy services assume connectivity and fail completely when it disappears.
✕ THE WRONG MODEL
Request / Response
App calls API
→ waits for network
→ gets entropy
→ proceeds
Every cryptographic operation pays a network tax. Minimum latency floor: 1 ms. Structurally irreducible regardless of throughput. Every existing EaaS is built on this model.
The Insight

The bottleneck is delivery, not production

If the server continuously pushes entropy into a client-resident buffer before the application needs it, the application never waits for the network.

✓ THE RIGHT MODEL
Persistent Server-Push Streaming
Server continuously pushes entropy → buffer
App reads from local buffer → ~µs
Network latency becomes buffer refill latency
Geography is irrelevant as long as refill > drain
THE RESULT Network-RTT-bound entropy access becomes memory-RTT-bound. No Western entropy service has built this.
Experimental Results

Demonstrated on NVIDIA H100 — fully reproducible

All benchmarks run on Lambda Labs H100 80GB HBM3. Code and raw results publicly available.

3.58 µs
Median Application Latency
32-byte session key read from local buffer
280×
Faster Than REST Delivery
vs. Qrypt best-case same-region API
2.71 → 7.93
H_min Amplification
bits/byte after Toeplitz extraction
4 / 5
Failure Modes Detected
Within 1 block (4 KB) — NIST SP 800-90B
|r| < 0.003
Inter-Channel Correlation
3 concurrent customers, shared pool
0 / 5000
False Positives
Health test FPR below NIST 2⁻²⁰ target
System Application Latency Notes
Qrypt REST API (cross-region) ~50 ms Typical enterprise, published cloud RTT
Qrypt REST API (same region) ~1 ms Best case — still 280× slower
ID Quantique Quantis PCIe ~15 µs Local PCIe driver call, hardware only
Thomes Quantum — streaming + buffer 3.58 µs Measured on H100, Python prototype
Service Model

Three delivery tiers

TIER 01
REST API
$500 – $2,500 / month
Unary request/response for occasional entropy needs. Easy integration for developers and SMBs.
  • Standard REST endpoint
  • Per-block health metadata
  • Ed25519 attestation
  • JSON or binary response
TIER 02
gRPC Streaming
$8,000 – $50,000 / month
Persistent server-push stream into client-resident buffer. Sub-microsecond application access. No per-call network overhead.
  • Persistent streaming connection
  • Local buffer, memory-speed reads
  • Per-customer channel isolation
  • Real-time H_min dashboard
  • Up to 60 concurrent channels / node
TIER 03
Sovereign / On-Prem
$500K – $5M / year
Air-gapped, pre-mission buffer loading for denied environments. Defense, classified, and forward operating base deployments.
  • Fully offline operation
  • Pre-mission buffer pre-load
  • Encrypted buffer at rest
  • Domestic supply chain
  • ATAK plugin integration
About

Built at the intersection of defense and quantum

Thomes Quantum was founded to solve the entropy infrastructure gap that post-quantum cryptography migration creates. The system is built on GPU-native distillation, validated against NIST SP 800-90B, and designed from the ground up for the latency and sovereignty requirements of the defense and financial sectors.

The full software prototype — CUDA kernels, gRPC streaming server, client SDK, and benchmark suite — is publicly available for independent verification.

github.com/rg-2006/qrng-distillation-paper
01
USAFA Cadet — Data Science + Quantum Technologies Sophomore studying Mathematical Modeling and Quantum Technologies at the United States Air Force Academy.
02
Top Secret Security Clearance Operational experience with CSEL radios and UAS systems in denied and contested environments.
03
Working Prototype — H100 GPU Full pipeline demonstrated: CUDA kernels, gRPC server, inline NIST health testing, per-block attestation. All results reproducible from public repository.
04
Technical Paper — arXiv cs.CR (pending) Peer-review-ready preprint with full benchmark methodology and reproducible experimental results.
Contact

Interested in a pilot or partnership?

Reach out to discuss Tier 2 enterprise streaming pilots, Tier 3 sovereign deployments, research collaborations, or investment inquiries.

Rob Gatt
Founder — Thomes Quantum
rob@thomesquantum.com