Modules
Five modules. One dataset.
The agents of the loop, split into products a fab can land one at a time. They share one fab-edge runtime, one SECS-GEM connector layer, one model router — and, critically, one compounding dataset.
The thesis
Each module makes the others more accurate
Bondix acts on the bonder using Warpix perception and Twinix pre-checks. Stackon sequences the stack using known-good-die data that Warpix produced. Twinix validates every move Bondix and Stackon propose. Yieldra scores all of it and pushes corrections back into each.
That is not a bundling argument. It is why the modules are worth more together than the sum of five point tools: they read and write the same corpus of bonding responses, defect imagery, molding outcomes, stacking records and yield telemetry.
Which means the honest way to buy is one module, proven on one workflow. The compounding is what makes the second and third obvious.
The set
What each module is for
- 01 Bondix — Sub-micron bonding, run by an agent.
Acts on the bonder using Warpix perception and Twinix pre-checks, then feeds every bond outcome back to Yieldra.
- 02 Warpix — See every void before it costs a package.
The eyes of the loop — X-ray/CT, SAM and AOI perception feeding Bondix, Stackon and Yieldra from one shared defect corpus.
- 03 Stackon — HBM stacks that ramp, not scrap.
Sequences stack, TSV and underfill using known-good-die data from Warpix, validated in Twinix before a single die is placed.
- 04 Yieldra — The yield engineer’s craft, encoded.
The brain — fuses defect, tool and test data from every agent into yield and ramp actions, and retrains the fleet.
- 05 Twinix — Hit the spec before the line runs.
The simulator — every Bondix, Stackon and Yieldra move is checked in a package-and-line twin before it executes.
Bondix
Sub-micron bonding, run by an agent
The acting module for die-attach, flip-chip and hybrid bonding.
What it does
Holds the bond, not just the setpoint
Bondix controls placement accuracy, bond force, thermal profile and sub-micron alignment in a closed loop, correcting drift as material lot, temperature and panel change. It consumes Warpix perception, pre-validates in Twinix, and refuses to commit a bond it cannot vouch for.
Pilot Bondix
Warpix
See every void before it costs a package
The perception module — the eyes of the whole loop.
What it does
One perception layer, five instruments
Warpix runs fine-tuned vision across X-ray and CT volumes, SAM maps and AOI images to detect voids, warpage, bump co-planarity and bridging, die cracks and delamination — and feeds Bondix, Stackon and Yieldra from one shared defect corpus rather than five disconnected threshold sets.
Pilot Warpix
Stackon
HBM stacks that ramp, not scrap
The sequencing module for 3D stacking, TSV and underfill.
What it does
Spends the good die on the good stack
Stackon optimises HBM and chiplet die-stacking, TSV planning and underfill sequencing using known-good-die data from Warpix, validated in Twinix before a die is placed — then hands the outcome to Yieldra to score. In a twelve-high stack, sequencing is not logistics. It is yield.
Pilot Stackon
Yieldra
The yield engineer’s craft, encoded
The decision module — and the memory of the line.
What it does
Turns four data streams into one decision
Yieldra fuses defect, tool, process and test data from every agent into yield and ramp actions — root cause with an owner, a DOE for tonight, a recipe change for this shift — and pushes retraining signal back to the rest of the fleet. It is where the fab’s accumulated craft is captured rather than lost to retirement.
Pilot Yieldra
Twinix
Hit the spec before the line runs
The simulation module — and the gate every other module passes through.
What it does
A move is simulated before it is made
Twinix runs a package-and-line twin that simulates bonding, stacking and molding, auto-optimising recipe and alignment against the target warpage, void and yield spec. Every Bondix, Stackon and Yieldra move is checked in the twin before it executes — which is what makes bounded autonomy defensible to a quality organisation.
Pilot Twinix
At a glance
Choosing where to land
Most fabs start with Warpix or Bondix. Both produce a number within a quarter.
| Module | Role in the loop | Typical wedge workflow | Primary metric |
|---|---|---|---|
| Bondix | Acts on the bonder | Hybrid-bonding alignment drift | Bond yield, alignment excursions |
| Warpix | Perceives the package | X-ray/CT void and warpage detection | Classification accuracy, escape rate |
| Stackon | Sequences the stack | HBM / chiplet stacking optimisation | Known-good-die loss, stack yield |
| Yieldra | Decides and retrains | Yield and ramp acceleration | Yield points, ramp weeks |
| Twinix | Simulates before the run | Pre-run recipe validation | First-pass spec hit rate |
Director of packaging operationsIntegrated device manufacturerEvery vendor sells us the tool. Nobody sells us the thing that makes six vendors’ tools behave like one line.
Composite drawn from design-partner and industry conversations. Illustrative, not a customer endorsement.
Shared foundations
What all five have in common
- One runtime The same fab-edge runtime, deployed once, hosting whichever modules you have licensed.
- One connector layer A single vendor-neutral SECS-GEM and API integration surface across the back end.
- One dataset Bonding responses, defect imagery, molding outcomes, stacking records and yield telemetry, compounding.
- One audit trail A single immutable record spanning every module, agent and approval.
Module questions
Buying and expanding
-
Yes. Each module is a complete product against its own workflow and metric. We would rather you expand because the first one worked than because a contract forced it.
-
No. The fab-edge runtime and connector layer are deployed once; additional modules are licensed and enabled on the existing footprint, subject to the sensor and compute headroom established at scoping.
-
Within your tenant, yes — that sharing is the point. Warpix perception improves Bondix control and Yieldra scoring. Across tenants, no, unless you have contracted for federated learning.
-
Whichever workflow has the clearest measurable pain. In practice that is usually X-ray/CT void detection with Warpix, or hybrid-bonding alignment control with Bondix, because both produce an unambiguous number inside a quarter.
Start narrow, expand relentlessly
Land one workflow. Own the loop.
A Chipira engagement begins with a single wedge workflow, a shadow-mode baseline and one signed success metric. Everything after that is expansion.