QuantHub by DXTECH
Request demo
QuantHub / by DXTECH

Decoding
Marketsinµs·ns

FPGA-accelerated market data and risk infrastructure for the microsecond economy. We decode, normalize, and risk-check every tick in hardware — at wire speed, in deterministic time.

Request a PoC See the architecture
FPGA-native pipeline KRX + global venues Built for KRW & KR regulation Two-way data gateway
QH-Feed · Live demo
Wire-speed
KRX + global venues
Tick-to-app < 1 µs
Wire-to-wire 480 ns
Risk check < 1 µs
Throughput wire-rate
1µs
Pre-Trade Risk
vs 100µs+ software baseline
480ns
FPGA pipeline
Deterministic, no GC jitter
<1µs
Tick-to-application
Sub-microsecond feed
7lines
Service portfolio
From feed to full appliance
The Problem

Software stacks have a ceiling. The market doesn't.

Algorithmic trading is now roughly 70% of US equity volume. The competitive unit has moved from milliseconds to microseconds to nanoseconds. Pure software stacks — with OS, GC and kernel jitter — can no longer keep up.

Jitter & non-determinism
OS scheduling, GC pauses and network stack cause unpredictable µs-level latency spikes.
Risk-check bottleneck
Software Pre-Trade Risk takes 100µs+ — a permanent ceiling on every low-latency order path.
Multi-venue normalization
Per-exchange protocols, symbols and time bases multiply development cost and add latency.
Cross-border friction
Both domestic firms reaching out and global firms reaching in face infrastructure and license walls.
Expensive global tooling
Proven FPGA stacks exist but are priced and shaped for the largest western funds — not mid-tier KR firms.
Architecture

Hardware on the hot path. Software on the cool path.

A hybrid design where every microsecond-sensitive step lives inside the FPGA fabric. Configuration, monitoring and operations stay in software — where flexibility matters more than nanoseconds.

Feed handler
Exchange ingest
~480 ns
Decode
Per-venue protocol
hw-pipelined
Normalize
Unified format
hw / sw
Pre-Trade Risk
Limits & checks
< 1 µs
Distribute
SHM / NIC / API
embedded
FPGA hot path · deterministic, wire-speed Software cool path · config, monitor, replay
The Hardware

Silicon-shaped for the order book.

QuantHub runs on FPGA fabrics tuned for the exact shape of market data: deep pipelines for decode and normalization, parallel risk-check engines, and an embedded path that writes ticks straight into your strategy's memory.

Kernel bypass, shared memory and FPGA-integrated NICs combine to remove every avoidable hop between the wire and the strategy. The result is a sub-microsecond, jitter-free path — not on average, but every time.

FPGA family
AMD / Xilinx UltraScale+
Form factor
1U appliance · PCIe card
Network
Dual 10 / 25 / 100 GbE
Determinism
Sub-µs, no GC, no OS jitter
Services

Seven lines, one infrastructure.

A complete portfolio — from raw exchange ingest to a turnkey appliance — covering every step of the low-latency value chain.

Lowest-latency market data

FPGA feed handlers deliver tick and trade data with deterministic, sub-microsecond performance. Embed mode writes directly into strategy memory.

Brokers · Prop · HFT

KR data to global desks

Normalized, low-latency delivery of KRX market & trade data to international trading firms and data vendors — a true two-way gateway.

Overseas firms · Vendors

FPGA custom builds

Bespoke FPGA logic and gateways shaped around your strategy, infrastructure and venues. End-to-end design, build and handover.

Exchanges · Fintech

FPGA Pre-Trade Risk

Wire-speed limit & order-suitability checks performed inline in the FPGA — sub-microsecond, deterministic, with zero impact on the hot path.

Featured · 1µs vs 100µs+

Pre / Post-Trade Risk · KR

Risk and compliance built for Korean regulation, exchange rules and KRW workflows — at a price the mid-tier can actually deploy.

Securities · Asset mgmt

Multi-venue normalization

One canonical schema across KRX and major global venues. Symbol mapping, timestamps and book state, unified.

Multi-market trading

Full integrated appliance

FPGA engine + Normalizer Server + Monitoring in a single, validated package. Plug it in and you have ingest, normalize, risk-check, distribute and observe — out of the box.

Featured · Flagship product
DXTECH Signature · Cross-Border Order Session HA

Order session HA — the part nobody else solves.

Market data HA is the easy half. The damaging failures happen on the order session — especially across KR ↔ overseas legs, where a 5-second flap can leave hundreds of orders in an unknown state. We turn that gap into a deterministic, two-mode recovery.

Market data
1 → N broadcast · stateless · idempotent
A/B feed arbitration solves it. Receive twice, pick the live one.
vs
Order session
N ↔ 1 stateful · non-idempotent · legally binding
A duplicate is a real trade. A lost ExecReport is a real exposure. The real question isn't "stay connected" — it's "know exactly where every in-flight order went."

Why the KR ↔ overseas leg is structurally fragile.

KR ↔ KRX is essentially solved. The damage concentrates on the cross-border legs — longer RTT, fewer carriers, multi-hop brokerage, and a recovery window measured in minutes, not milliseconds.

Leg
RTT
Carrier diversity
Incidents / quarter
State resync after reconnect
KR ↔ KRX (EXTURE+)
< 1 ms
Fully controlled
≈ 0
~ ms
KR ↔ Japan / Hong Kong
30 – 80 ms
2 – 3 subsea cables
1 – 2 / quarter
sec – min
KR ↔ United States
120 – 150 ms
Cables + intermediate brokers
1 – several / month
min – tens of min

Why 1 minute is the automation / human-intervention boundary.

The threshold isn't arbitrary — it's where five separate timing events all converge. Before 1 minute, automated recovery is still tractable. After, market movement, venue-side cancels, and operator engagement all activate at once.

FIX Heartbeat miss (30s config)
~ 30 s
FIX Session strong disconnect
~ 60 s (2 missed heartbeats)
Venue-side Cancel-on-Disconnect (COD) trigger
typically 30 – 60 s
Material market movement (volatility regime)
~ minutes
Operator detection + engagement complete
~ minutes
1 min = the last point at which automated recovery is still the right answer.
Mode 1 · < 1 min
Auto-Heal
"Where did every in-flight order land?"
  • ·Hot-Hot FIX sessions (Primary/Backup logged in)
  • ·Fast heartbeat 1–3s, sub-100ms failover
  • ·In-Flight Tracker + WAL · idempotent retransmit by ClOrdID
  • ·Auto-reconciliation via Drop Copy (zero operator touch)
Detect→switch< 1 s
Auto-recovery99.9%+
User impactnone
Mode 2 · 1 – 10 min
Controlled Recovery
"How do we hedge the exposure right now?"
  • ·Cancel-on-Disconnect (COD) at 60–90s · exposure capped
  • ·Hedge Router → alternative venue / index futures / put options
  • ·Drop Copy → position truth source even while session is down
  • ·Operator dashboard · 1-click ACK · auto venue escalation
Exposure window< 90 s
Position accuracy99.99%
Operator decide< 30 s

The two modes, line for line.

Same appliance, two operating modes — each with its own central question, automation ratio, COD policy, and target SLO. The cross-over is automatic at the ~1 minute mark.

Dimension
Mode 1 · Auto-Heal
Mode 2 · Controlled Recovery
Underlying issue
Transient line / session glitch
Market exposure · position risk
Central question
"Where did every in-flight order land?"
"How do we hedge the exposed position now?"
Automation ratio
100%
80% automated + 20% operator ACK
Critical component
Hot-Hot session + In-Flight Tracker
Drop Copy + Hedge Router + Dashboard
COD policy
Disabled — protect short flaps
Enabled — 90s threshold caps exposure
Operator notification
Post-incident report only
Immediate alert · dashboard entry
Target RTO
< 1 s
< 90 s
Target RPO
0 (idempotent retransmit)
0 (Drop Copy + reconciliation)
Analogy
Lane change (driver doesn't notice)
Emergency stop (decision required)
01
A/B FIX Session Multiplexer
Dual sessions on diverse carriers. Same pattern as our A/B market data feed — direct port of a proven asset.
02
Order State Reconciliation
Per-order reconcile of our WAL against exchange Drop Copy — automatic resolution of every state pairing, operator escalation only on true conflicts.
03
FPGA Pre-Trade Risk Inline
Hedge orders go through the same 1µs risk check as primary flow. No emergency back door, no widened limits.
04
Hedge Router (Mode 2)
Rule-driven hedge engine: direct cross-venue → index futures → puts → inverse ETF, prioritized by health-checked routes. Pre-agreed thresholds, one-click ACK.

The five failure patterns that actually happen on the floor.

Patterns 1, 2 and 4 are the ones that surface to retail customers as "order status unknown" — the image-damage scenarios. We engineer against each pattern explicitly.

01
Line flap (5-second disconnect)
FIX heartbeat (typically 30s) is longer than the flap, so the gap goes undetected. Orders sent in that window end up in an unknown state.
02
One-way break (order out, no ExecReport)
Order leaves the gate but the execution report never returns. Was it filled? Rejected? You don't know.
03
Sequence gap after reconnect
Resend Request is required, but the two sides disagree on which sequence numbers they hold.
04
Duplicate orders
Operator assumes the session is dead and re-sends — but the original went through. Both fill. Real exposure, real settlement.
05
Overseas-side reject
Korean overnight is overseas midday — peak volatility window with no Korean operator on shift.

Eight techniques. They only work when applied together.

The reference set used by Activ, Fidessa, ION, TT. Any single technique on its own is a half-fix. The gateway implements all eight on one box.

01
Dual FIX sessions
Single-line failure
Medium
02
Short heartbeat + fast failover
Flap-detection latency
Low
03
Cancel-on-Disconnect (COD)
Auto-cancel in-flight on session loss
Venue-dependent
04
Drop Copy feed
Single source of truth
Separate venue contract
05
Order State Reconciliation Engine
Post-recovery state alignment — the core problem
High
06
WAL + persistent store
Restart-time state recovery
Medium
07
Pre-allocated ClOrdID + idempotency
Duplicate prevention on retransmit
Low (design discipline)
08
Smart router + line health-check
Never route into a dead line
Medium

The recovery moment is a pairwise decision matrix.

On reconnect, every in-flight order is reconciled against the exchange's Drop Copy view. Each pairing has exactly one correct action — the engine resolves it, no operator needed.

Our state (WAL)
Exchange (Drop Copy)
Resolution
PendingNew
(none)
Not received. Safe to retransmit.
PendingNew
New
Accepted. Do not retransmit.
PendingNew
Filled
Already filled. Begin hedge.
New
Filled
Fill notification lost. Replay ExecReport.
PartialFill (5/10)
Filled (10/10)
Remainder fill lost. Replay residual ExecReport.
(none)
Filled
Our WAL is missing the order. Escalate — this is the near-miss case.

Why we're not building from scratch — direct asset reuse.

The order gateway is not a parallel codebase. It's a re-application of the assets we already ship and operate — same FPGA, same normalization vocabulary, same monitoring console.

Existing asset (market data)
→
Order gateway application
DXTECH SBE Profile (normalization)
→
Cross-venue order response normalization · unified Drop Copy view
FPGA decoder / SDK
→
Pre-trade risk in FPGA · 1µs limit check
A/B dual gateway pattern
→
A/B FIX session multiplexer · identical topology
Gap detection + reconciliation
→
Order sequence gap detection · same logic
Appliance form factor
→
1U DXTECH Order Gateway appliance
Bilingual (KO/EN) monitoring console
→
24/7 global operations console

DXTECH Cross-Border Order Gateway — one box, both flows.

The eight techniques and the reconciliation matrix, packaged as a 1U appliance that sits between the broker OMS and the overseas DMA. Four differentiators set it apart from a software-only stack.

Broker OMS
KR side
FIX
DXTECH Order Gateway
1U appliance · diverse-carrier dual line
  • ·A/B FIX session multiplexer
  • ·Order state reconciliation engine
  • ·FPGA pre-trade risk
  • ·WAL + Drop Copy aggregation
  • ·COD orchestration
↓
Real-time alerts → operations console + SLA report
FIX
Overseas broker / DMA
NYSE · HKEX · SGX · TSE
Differentiator 01
FPGA pre-trade risk
Limit check in 1µs. Software-only stacks land at 100+ µs.
Differentiator 02
Cross-venue normalization
Same SBE-style approach as our market data — absorbs NYSE / HKEX / SGX / TSE quirks.
Differentiator 03
Automated reconciliation engine
Operator headcount stays flat as venues grow.
Differentiator 04
Same appliance for both flows
Market data and order gateway in one box — single TCO line, single ops surface.
~60–70% of our market-data assets — A/B routing, gap detection, normalization, FPGA, monitoring — port directly into the order gateway. One appliance, both flows.
Pre-Trade Risk

1 microsecond. Not 100.

Pre-Trade Risk is non-negotiable: every order must pass position, notional, price and duplicate checks before it leaves the firm. Software implementations make this the slowest step in the path. We move it into silicon.

Software baseline
≥100µs
A permanent ceiling on every order.
  • Jittery, non-deterministic latency
  • Bottleneck on every low-latency strategy
  • Risk lives outside the wire path
  • Order flow waits for the software stack
QuantHub FPGA
< 1µs
Risk lives on the wire, not behind it.
  • Deterministic, wire-speed
  • 100× faster than software baseline
  • Inline FPGA logic — zero hot-path cost
  • Built for KR limits, rules and workflows
Flagship Appliance

FPGA + Normalizer + Monitoring. One box.

Our flagship product is a single, validated appliance. Ingest, normalize, risk-check, distribute and observe — from a single SKU, with a single support contract.

FPGA acceleration engine
Decode / Normalize / Pre-Trade Risk
Normalizer Server
Multi-venue normalization · symbol mapping · API
Monitoring layer
Latency · throughput · risk limits · alerts
Customers

Built for the desks that can't wait.

From mid-tier securities firms to global prop shops needing access to KRX data — QuantHub fits the desks where latency, regulation and cost all matter at once.

Securities & futures brokers
DMA / algo · internal-control Pre-Trade Risk
→ 1µs risk + full appliance
Prop trading & HFT
Lowest latency feed · deterministic performance
→ FPGA feed handler + embed
Asset managers & hedge funds
Multi-market data · post-trade risk
→ Normalization + Post-Trade
Exchanges & fintech
Data infrastructure · custom builds
→ Custom build + appliance
Overseas trading & vendors
Low-latency access to KR data
→ KR data → global gateway
Competitive Landscape

Eight dimensions. One stack that covers them all.

QuantHub is the only domestic player combining FPGA-grade performance, KR-fit risk and regulation, and a fully integrated appliance. Competitors cover parts of the picture; we cover all of it.

Capability
QuantHub
Competitor A (KR)
Competitor B (Global)
Major Global Vendor
Software Specialist
FPGA-based
core
Domestic exchange integration
KR-fit risk & regulation
KR data to overseas
Multi-venue normalization
TCO efficiency
High
Medium
Low
Low
Medium
Full integrated appliance
Local support & response
Full coverage Partial Not covered

QuantHub is the only domestic player to combine Korean-regulation risk capability with FPGA performance — at materially lower TCO and with fast local support compared to global vendors.

Positioning

Three references. One combined play.

We combine FPGA architecture from a proven global vendor, domestic credibility from an established Korean integrator, and the long-term subscription shape of a leading data SaaS — unified on Korean ground.

Dimension
Competitor A (KR)
Competitor B (Global)
Competitor C (Global)
QuantHub
Core area
Low-latency network · FEP
FPGA feed · Pre-Trade Risk
Data · Risk SaaS
FPGA feed + KR Risk + Normalize
Model
Build · hardware · maintenance
Appliance · license
SaaS subscription
Appliance + SI + subscription
Strength
Korean broker references
Proven FPGA performance
$10T AUM · 2,400 clients
FPGA performance · KR-fit · price
Gap we close
Limited FPGA / normalization
Cost · localization
Not ultra-low-latency focused
All three, in one local stack
Vision

Decoding markets, from Seoul to the world.

Vision

Asia's leading FPGA infrastructure company

Become the standard for ultra-low-latency market data and risk infrastructure across Asia's capital markets.

Mission

Fast · Accurate · Affordable

Decode market data as quickly, accurately and affordably as possible — so every desk, not just the largest, can compete at the microsecond.

Slogan

Decoding Markets

Translate the chaos of raw exchange feeds into clean, fast, actionable signal — for the people who trade on it.

Speed
Deterministic µs / ns performance
Accuracy
Unified normalization, one schema
Affordability
Built so mid-tier can deploy
Integration
FPGA + Normalizer + Monitoring
Global
Two-way KR ↔ world gateway

Bring QuantHub to your desk.

We're partnering with a small number of Korean and global desks on Phase 1 proofs of concept. If you operate in microseconds, we'd like to talk.

Request a PoC Talk to engineering