End-to-end physics twin
Paste deposition, placement, reflow, and joint formation.
Digital Twin · Circuex agent 04
Twinex
The SMT-line digital twin that hits target yield first. Simulates paste, placement, reflow, and joint formation end-to-end, then hands a qualified recipe to the line before the first board runs.
Simulate & proveGraduated autonomyMES write-back
DIGITAL TWIN · NPI
98.7%Predicted FPY
60%Faster NPI
0Boards burned
AUTOMOTIVEMEDICALAEROSPACEINDUSTRIALAI HARDWAREEMS / CM
The problem
Why this stage bleeds
NPI and changeovers are expensive gambles: engineers pick settings from experience, run first articles, inspect, and iterate — burning boards, machine time, and days per program while scrap piles up.
The solution
How Twinex closes it
Twinex simulates the physical line end-to-end and auto-optimizes recipes to a target yield, validating every Stencix/Reflux move against the twin and continuously reconciling against real Voidex/Yieldix outcomes.
Capabilities
What Twinex does
Pre-run optimization
Recipes auto-tuned to a target first-pass yield.
Synthetic rare defects
Tombstones, BGA voids, bridges, and head-in-pillow stress-tested.
Process validation gate
Stencix/Reflux moves proven in-twin before production.
Continuous calibration
Reconciled against real Voidex/Yieldix outcomes.
What-if planning
Changeover and substitution simulation for high-mix.
DIGITAL TWIN · NPI
98.7%Predicted FPY
60%Faster NPI
0Boards burned
In the loop
Twinex never works blind
It shares perception, twin validation, and orchestration with the rest of Circuex — so every action is grounded and every result feeds the next board.
- Simulate & prove — its role in the perceive–simulate–decide–act loop.
- Twin-validated — moves proven in Twinex before production.
- Cited & logged — assurance-grade audit trail on every change.
Autonomy
Graduated, never reckless
You choose how much control Twinex takes — and it earns the next level by proving itself.
01
Shadow
Twinex perceives and predicts with zero write-back — you validate its calls against reality.
02
Recommend
It surfaces grounded recommendations with the evidence behind each one.
03
Approve
Operators approve actions one at a time under policy, building trust and data.
04
Closed-loop
Twinex tunes autonomously within guardrails, escalating only exceptions.
Outcomes
What Twinex moves
0First-pass yieldin the loop
0Fewer false callsvs. rule-based
0Less scrap & rework
0Edge inferenceper tile
Built on NVIDIA
Technology overview
The physical-AI stack behind Twinex
Trained on DGX/HGX, served at the edge on Jetson and IGX, validated in Omniverse and Isaac Sim.
JetsonEdge inference
TensorRTLow-latency serving
OmniverseTwin & synthesis
DIGITAL TWIN · EDGE
20–50msPer tile
8–16Cameras
LiveOn line
Once Twinex went closed-loop, the stage stopped being the bottleneck. It just holds yield — and tells us why when it doesn't.
Questions
Twinex FAQ
Does Twinex require the other agents?
No. It's a standalone product with its own ROI, but it compounds when paired with the rest of the Circuex loop.
How does it connect to our equipment?
Read-only connectors first, then graduated write-back to the relevant machines and your MES/traceability system.
Is it safe for high-reliability production?
Yes — graduated autonomy, approval policies, and an assurance-grade audit trail are built in for automotive, medical, and aerospace.
Every board, perfectly built.
See Twinex run on your boards
Book a working session and watch Twinex perceive, predict, and close the loop against your line data.