CoreTech System Co., Ltd. (brand: Moldex3D)¶

Company Overview¶

Business Overview¶

CoreTech System Co., Ltd. was founded in November 1995 in Taiwan by Professor Rong-Yeu Chang (National Tsing Hua University) and four co-founders: Dr. Chia-Hsiang (David) Hsu, Dr. Wen-Li (Venny) Yang, Mr. Ming-Hong (Vito) Tsai and Dr. Wen-Hsien (Anthony) Yang. It is headquartered in Zhubei City, Hsinchu County, Taiwan. [Evidence] ¹⁹²⁰

Key corporate facts:

Ownership: private. Chinese-language unlisted-stock trackers state plainly that CoreTech is not listed on Taiwan's main board, OTC board (上櫃) or emerging-stock board (興櫃). Consequently no audited public financials exist; revenue figures in this report are absent by necessity, not oversight. [Evidence] ²¹
Size: roughly 192 employees per a business-data aggregator — a secondary source; treat as approximate. For scale: this is ~1% of Dassault's headcount, yet the company holds its own against Autodesk Moldflow. [Evidence — secondary] ³¹
Market position: Chinese-language business press describes CoreTech as "Asia's largest and the world's second-largest" injection-moulding CAE company. This is consistent with the industry's common framing of Moldex3D and Autodesk Moldflow as the two dominant independent mould-flow solvers, but no independent market-share audit was found. [Evidence — secondary] ²⁰²¹
Leadership: a generational transition took effect April 19, 2022 — founder Rong-Yeu Chang became Honorary Chairman and Chief R&D Consultant; Dr. David Hsu became Chairman and Dr. Venny Yang became CEO (both co-founders). [Evidence] ²²

Automotive Business¶

Moldex3D does not publish an automotive revenue split (as a private company it publishes no revenue at all). Automotive is one of its named industry solution areas, and its public case-study library is rich in automotive suppliers — window encapsulations, car-roof parts, optical lenses, mould makers (see §7). [Evidence] ²⁷²⁸³⁰

Engineering Business¶

The whole company is one engineering niche: simulating how molten plastic fills, packs, cools and warps in an injection mould — "mould-flow" simulation, i.e. moulding CAE (one branch of computer-aided engineering) — plus the data and services around it — a materials-testing laboratory (ISO 17025 accredited), machine characterization, training, and since 2023 a cloud ecosystem called Moldiverse. [Evidence] ²⁴²³

Role in Automotive Moulded Parts Workflow¶

Workflow Stage	Relevant?	Notes (product)
Requirements	No	Not a PLM/requirements vendor
Industrial Design	No	No CAD authoring of its own
Concept Design	Partial	Early mouldability feedback via EasyFill / DesignSim in NX
CAD Modelling	No (embeds into others' CAD)	SYNC for NX/Creo; no native CAD
Engineering Review	Partial	iSLM stores and compares simulation + tryout records
Simulation	Core strength (moulding only)	The Moldex3D solver; no structural/CFD ambitions
DFM	Yes — rules-based	DesignSim (Jul 2026): draft, undercuts, sink marks, L/t ratio
Tool Design	Partial	Cooling Channel Designer; gate/runner design aids; not full tool CAD
Mould Flow	Core strength — the product	Fill/pack/cool/warp, fibers, multi-component, potting
Prototype	Partial	Virtual mould trials (Molding Window Advisor)
Validation	Yes	Simulation vs. mould-trial records in iSLM
Manufacturing Engineering	Partial	iMolding Hub: machine characterization, trial-moulding guidance
Production Release	No	Not a PLM vendor

What makes Moldex3D structurally different: it is narrow and deep. It covers only the moulding-relevant stages — but it covers them as its entire existence, and it is embedded inside the platforms of vendors who cover everything else. [Synthesis]

AI Strategy¶

Public AI Vision¶

Moldex3D's AI story is branded release by release rather than as a grand corporate strategy:

Moldex3D 2025 (March 26, 2025) was themed "Molding Intelligence" — "automation and AI-powered collaboration, providing intelligent recommendations from process parameter settings to gate and mold design optimization." [Evidence + Marketing] ¹
Moldex3D 2026 (March 20, 2026) organised everything under "A.O.I." — Automation, Optimization, Intelligence — described as the release's "three major development pillars." Note: A.O.I. is a framing acronym, not a product. [Evidence] ²⁷

The underlying pitch is consistent: convert thirty years of moulding simulation know-how — solver, measured material data, and customers' accumulated project data — into recommendations, so that fewer physical mould trials are needed. [Synthesis]

Investor Statements¶

None exist — CoreTech is private and publishes no investor communications. This is itself a data point: every AI claim must be assessed from product documentation and trade press, with no financial-disclosure discipline behind it. [Evidence of absence] ²¹

Engineering AI Strategy¶

Read across the 2024–2026 announcements, the actual technical strands are: [Synthesis]

Retrieval over accumulated project data — iSLM's similarity search ("Discovery" series) over a company's own historical mould designs and results. Case-based reasoning, not model training.
Bought-in LLM + RAG — MoldiBot "uses ChatGPT as its core language model" with retrieval-augmented generation over Moldex3D's courses, documents and videos. The LLM is OpenAI's, not CoreTech's. [Evidence] ⁴⁵
Simulation-trained prediction models — Process Discovery claims "deep learning data science methods" building an "AI prediction model" from batches of the user's own simulation runs — i.e. a surrogate model (see §5.4). [Evidence — thin] ⁴
Classical optimization and DoE, relabelled — the AI Optimization Wizard and DOE Wizard automate parameter search across simulation runs. [Evidence + Marketing]³³²

Timeline of AI Evolution¶

Date	Event
Jan 2010	Joins Siemens PLM Software & Technology partner program¹⁴
Dec 2012	Moldex3D solver embedded as Siemens NX EasyFill Analysis¹³
Oct 2013	US patent 8,571,828 — fiber-orientation model (physics, pre-AI context)²⁵
~2018–2020	iSLM launched as "big-data management platform" for moulding¹⁰
2023	Moldiverse cloud ecosystem launched (Material Hub Cloud, iMolding Hub…)²³
Dec 16, 2024	MoldiBot announced — ChatGPT-based assistant on Moldiverse⁵
Mar 26, 2025	Moldex3D 2025 "Molding Intelligence" — AI Optimization Wizard; iSLM Mold/Gate Design Discovery; MoldiBot GA¹
Apr 22, 2025	SPE Injection Molding Division technical write-up details mechanisms (RAG, LLM cooling channels, Process Discovery)⁴
Apr 2025	"Best of ANTEC 2025" award — automotive electronics potting simulation (not ML)²⁶
Mar 20, 2026	Moldex3D 2026 "A.O.I." — Moldibot "Deep Thinking", AI Chat, AI Material Assistant, Gate Location Advisor, DOE Wizard constraints²
Jul 16, 2026	DesignSim launched — rule-based DFM validation inside Siemens NX¹²¹¹

Products Relevant to Engineering¶

Moldex3D (Studio) — the flagship solver. Purpose: true-3D simulation of filling, packing, cooling, warpage, fiber orientation, multi-component moulding, potting and more. AI: the 2025/2026 "intelligence" features listed in §5 sit on top; the solver itself remains classical numerical physics. Public evidence: release notes 2025/2026; decades of case studies. [Evidence] ¹²

iSLM — the data/knowledge platform. Purpose: centralise CAD models, simulation records, mould-trial results and process settings; make them searchable ("Knowledge Management" filters by industry, product, mould properties). AI: the Discovery series (2025–26) adds geometry-similarity retrieval and predictive insights from matched historical cases; 2026 adds AI Chat natural-language queries. This is the closest thing to a shipping "moulding knowledge base" found anywhere in this study. [Evidence] ⁹⁸²

Moldiverse — the cloud ecosystem (2023). Purpose: Material Hub Cloud (8,000–9,000+ measured material grades), iMolding Hub (machine characterization, trial-moulding guidance), University (training), Store, Forum. AI: hosts MoldiBot; 2026 adds an "AI Material Assistant" on Material Hub Cloud (thinly described). [Evidence] ²³²⁴²

DesignSim (July 16, 2026) — DFM validation inside Siemens NX. Purpose: designers check mouldability without leaving CAD: undercuts, draft angles, sink-mark prediction, flow-length-to-thickness ratio, cooling-time estimate, filling simulation, weld lines/air traps; "scores molding feasibility using a rule-based DFM analysis" with automated solver defaults. Vendor claims up to 4.75× faster validation vs. conventional CAE workflows. AI: explicitly not AI — rules plus the real solver. [Evidence] ¹¹¹²

CAD plug-ins — SYNC for NX and PTC Creo (formerly eDesignSYNC), Moldex3D eXplorer for NX, and the Siemens-branded NX EasyFill / Easy Fill Advanced powered by Moldex3D. [Evidence] ¹⁵¹⁶¹⁴¹³

AI Capabilities¶

Each capability below is held against the study's honest test: does it learn from data, or compute from equations (or look things up)?

5.1 AI Optimization Wizard (2025)¶

Description: "By simply providing parameter ranges, it can automatically recommend results that meet the requirements"; "AI dynamically adjusts the parameter settings to quickly find the best solution"; supports single- and multi-objective problems; outputs parallel-coordinate plots and CSV. [Evidence + Marketing] ¹³⁴ Verdict: No disclosed learning mechanism. The described behaviour — search a user-supplied parameter range by running simulations, present a trade-off plot — is the classic pattern of automated design-of-experiments / iterative optimization, the direct successor of Moldex3D's long-standing "Expert" DoE module. Until CoreTech documents a trained model, this is classical optimization wearing the AI label³². See the concept note: What is Optimization technology. Workflow stage: process-parameter setup. Limitation: mechanism undisclosed; no accuracy claims; every candidate still costs a solver run. [Synthesis]

5.2 DOE Wizard (2026)¶

Description: enhanced in 2026 with "Design Constraints" to balance quality factors against production constraints. [Evidence] ² Verdict: explicitly named design-of-experiments — classical virtual DoE, honestly labelled. Its coexistence with the "AI Optimization Wizard" strengthens the reading that the latter is the same family of technology with different branding³². [Inference]

5.3 Molding Window Advisor (2025)¶

Description: automatically sets up systematic "CAE scientific trial molding" runs and identifies suitable moulding parameters; delivered via the Moldiverse cloud, usable from a mobile browser at the machine. [Evidence] ³⁴ Verdict: automated virtual mould trials — simulation automation, not learning. Genuinely useful; not AI. [Synthesis]

5.4 Process Discovery (2025) — the strongest real-ML claim¶

Description: "combines the Moldex3D analysis project results from Molding Window Advisor and uses deep learning data science methods to build an AI prediction model" that "provides real-time prediction capabilities, so when molding conditions change, it can quickly generate molding results as references for machine adjustments." The model can be uploaded to iMolding Hub to guide mould trials from a mobile browser. [Evidence] ⁴ Verdict: this is a genuine machine-learning surrogate claim in the moulding domain — a neural network trained on a batch of the user's own simulation runs, then predicting moulding results instantly as process settings change. Two honest caveats: (a) it learns from simulation outputs, not measured factory outcomes — the physics is still the teacher; (b) documentation is thin — no architecture, accuracy figures or benchmark has been published. [Evidence — thin; Synthesis] Workflow stage: process setup and mould-trial support.

5.5 iSLM Discovery series (2025–2026) — retrieval, shipping today¶

Description: upload a CAD model; the system "performs similarity comparisons against historical projects stored in the Knowledge Base," analysing "the part's geometry, dimensions, and structure to identify the most relevant reference cases," then combines database records (materials, mould design parameters) into predictive insights — sprue pressure, clamping force, shrinkage, cycle time. Gate Design Discovery generates a ranked list of high-similarity gate designs; Mold Design Discovery estimates gate counts, machine selection, injection pressure and clamping-force distribution probabilities. The vendor stresses transparency: iSLM "does not directly apply optimized results" but "indicates the projects and items referenced." Marketing calls the matching "AI shape recognition." 2026 extends the series to proactive defect flagging. [Evidence] ⁸⁴² Verdict: case-based retrieval, not model training — geometric similarity search over a company's own accumulated projects. By the honest test this is "looking things up," not learning — but it is exactly the retrieval-knowledge-base concept this study's own research identifies as the high-ROI, label-free win, and Moldex3D is the first vendor found shipping it for moulding. How the shape similarity is computed (learned embeddings vs. handcrafted descriptors) is not disclosed. [Synthesis] Limitation: value scales with how much historical data a customer has loaded into iSLM; useless on day one.

5.6 MoldiBot (announced Dec 2024; evolved 2026)¶

Description: a 24/7 assistant on Moldiverse, "built on ChatGPT technology," using retrieval-augmented generation (RAG) over Moldex3D's courses, technical documents and tutorial videos; answers software and moulding questions in multiple languages with document links. Moldex3D 2026 adds a professional knowledge base, forum content and "Deep Thinking" for complex problems. [Evidence] ⁵⁴² Verdict: a real LLM application, honestly generic. The language model is OpenAI's, bought in; CoreTech's contribution is the domain corpus. It is a support/documentation assistant — it does not read your simulation results or design your part. [Synthesis]

5.7 AI Chat in iSLM (2026)¶

Description: natural-language diagnostics and data retrieval over the iSLM database. [Evidence — thin] ² Verdict: LLM front-end on the knowledge platform; mechanism details absent.

5.8 Generative-AI cooling-channel generation (announced 2025)¶

Description: "based on Large Language Models (LLM), uses input from product and mold descriptions and CAD files as prompts, automatically generating cooling channel design suggestions," with final designs "optimized and verified using Moldex3D." [Evidence] ⁴⁶ Verdict: if real, this is generative AI touching tool design — a first. But it was announced as a technology development, and no shipping product, product page, demo or benchmark could be found. Treat as a lab capability until proven. [Evidence — thin]

5.9 AI Material Assistant (2026)¶

Description: on Material Hub Cloud, "transforming simulation data into intelligent assets." [Evidence — thin] ² Verdict: too thinly described to classify. Appendix C.

5.10 DesignSim DFM (Jul 2026) — explicitly not AI¶

Description: rule-based mouldability scoring (undercuts, draft, sink marks, flow-length/thickness, cooling time) plus real filling simulation, inside NX; "up to 4.75×" faster validation [Marketing]. [Evidence] ¹¹¹² Verdict: rules + solver — the vendor's own description says "rule-based DFM analysis." Same category as SOLIDWORKS DFMXpress, but backed by a far better solver. Not AI, and not claimed to be — creditable honesty.

5.11 Gate Location Advisor (2026)¶

Description: automatically suggests gate positions, reducing manual iteration. [Evidence — thin] ² Verdict: mechanism undisclosed; gate-location advisers are traditionally flow-resistance heuristics. Distinct from iSLM's Gate Design Discovery (which is retrieval-based). Assume heuristic until documented otherwise. [Inference]

Engineering Workflow Contribution¶

Moldex3D does not carry a part end-to-end; it saturates the moulding-critical middle of the workflow, usually from inside someone else's CAD:

part designed in NX/Creo (or CATIA, exported) → early mouldability check with DesignSim / NX EasyFill inside CAD → full Moldex3D fill/pack/cool/warp simulation → parameter search via DOE / AI Optimization Wizard / Molding Window Advisor → gate & mould design informed by iSLM Discovery retrieval from past projects → mould trial guided by Process Discovery predictions on iMolding Hub → simulation + trial records archived back into iSLM as searchable knowledge.

The last step matters most for this study: Moldex3D is the only vendor studied whose product loop explicitly accumulates a company's moulding history into a queryable knowledge base. [Synthesis] ⁹⁸

Public Customer Evidence¶

All cases below are vendor-published customer stories: outcomes are self-reported and unaudited — [Evidence + Marketing] throughout. Note also that every quantified automotive case found predates the 2025–26 AI features: they evidence the solver, not the AI.

Case Studies¶

DURA Automotive Systems (India) — vehicle window encapsulation (PVC over-moulding with glass and aluminium insert). Moldex3D analysis located the root cause in a thick PVC section; an ABS insert fixed it. Reported outcomes: X-deflection 5 mm → 1 mm; cooling time 112 s → 38 s; sink marks eliminated. Published October 3, 2019. The strongest quantified automotive case found. [Evidence + Marketing] ²⁷
Car roof cartridge part — warpage optimization using "pre-warp" (machining the cavity with compensating deformation): total warpage reportedly reduced from ~18 mm to ~3 mm. [Evidence + Marketing] ²⁸
"World's leading automotive part supplier" — validation case: warpage predictions "proven highly reliable" against injection trials. The customer is not named in the title but is now confirmed as Faurecia Interior Systems India Pvt. Ltd. (Pune) — a direct fetch of the case page (this study's Forvia research, 2026-08-02) shows the body names Faurecia explicitly: centre-console fascia, tool-tuning cost −68%, weld lines relocated (see the Forvia report §7). [Evidence + Marketing — customer confirmed 2026-08-02] ²⁹
Frumolde (Portugal)³³ — automotive mould maker; testimonial on Moldex3D integration in mould design/manufacture, including a pre-machined warpage compensation case producing a part with "zero warpage." No independent verification. [Evidence + Marketing] ³⁰
Chao Long Motor Parts Corp. — joint ANTEC 2025 paper with CoreTech on the potting process of an electronic component (jet-ski application), awarded "Best of ANTEC 2025." Peer-reviewed venue, but the content is simulation of potting, not AI. [Evidence] ²⁶

Conference Demonstrations¶

Annual "What's New" webinars for 2025/2026 releases; ANTEC 2025 paper award. [Evidence] ¹²⁶

White Papers / Testimonials¶

The customer-success blog is extensive (hundreds of entries). A systematic gap: no public case study yet demonstrates the 2025–26 AI features (Discovery, Process Discovery, MoldiBot) delivering a quantified customer outcome. [Evidence of absence]

Technical Architecture (Inferred)¶

Evidence¶

Desktop solver (Moldex3D Studio) with a Python/API automation layer.³
Moldiverse cloud: Material Hub Cloud (ISO 17025 lab-measured data), iMolding Hub (machine characterization + trial guidance), University, Store, Forum; single-account access.²³²⁴
iSLM: web-based data-management platform holding CAD, simulation, tryout and process records; similarity search and (2026) AI Chat on top.⁹²
MoldiBot: ChatGPT as core LLM + RAG over Moldex3D's corpus.⁴⁵

Synthesis¶

The stack reads: solver (desktop/HPC) → Moldiverse cloud services (materials, machines, training) → iSLM as the customer-specific knowledge layer → LLM assistants (MoldiBot, AI Chat) as the interface layer. The LLM is bought in (OpenAI) — structurally the same dependency pattern as Dassault–Mistral.

Inference¶

Process Discovery models are almost certainly per-project surrogates (trained on that project's simulation batch), not a global foundation model of moulding — nothing suggests cross-customer learning, and iSLM's design (each company's own knowledge base) points the same way. How iSLM's "AI shape recognition" computes similarity — learned 3D embeddings vs. engineered geometric descriptors — is undisclosed; either would work, and the vendor's own documentation avoids claiming deep learning for it. [Inference]

AI Technologies¶

In one list: an LLM chatbot with RAG (MoldiBot — ChatGPT, bought in, not owned); natural-language data query (AI Chat); geometry-similarity retrieval over historical projects (iSLM Discovery — mechanism undisclosed); claimed deep-learning surrogate models trained on simulation batches (Process Discovery); announced LLM-driven generative design for cooling channels (unshipped, thin); classical DoE and automated optimization (DOE Wizard, AI Optimization Wizard³²); rule-based DFM (DesignSim); digital-twin elements via machine characterization (iMolding Hub). [Evidence + Synthesis]

Research Publications¶

Papers¶

"Best of ANTEC 2025" award paper with Chao Long Motor Parts on electronics potting simulation — peer-reviewed venue; classical simulation, not ML. [Evidence] ²⁶
The academic literature is full of third-party papers that pair Moldex3D simulations with neural networks (warpage ANNs, transfer learning, PINNs) — evidence that Moldex3D is the standard data generator for moulding-ML research, but these are not CoreTech's own publications and are not claimed as such here. [Synthesis]

Patents¶

US 8,571,828 (announced Oct 2013) — the iARD-RPR fiber-orientation model. Physics, not AI; included as context for the company's R&D character. [Evidence] ²⁵
A US patent, 11,220,033 ("mold model for production predictive control"), surfaced in searches with plausible relevance, but its assignee could not be verified as CoreTech from the sources retrieved — listed in Appendix C, not claimed. [Unverified]

Standards¶

None found specific to AI. The materials lab's ISO 17025 accreditation is a measurement-quality standard, not AI. [Evidence] ²⁴

Open Source¶

No verified open-source AI footprint. [Evidence of absence]

No official CoreTech System / Moldex3D GitHub organisation or Hugging Face presence could be verified. (GitHub organisations named "CoreTech" exist but belong to unrelated companies of the same common name.)
The Moldex3D API with Python scripting enables workflow automation but is a proprietary product feature, not open source. [Evidence] ³
Consistent with the company's character: proprietary solver, proprietary measured material data, proprietary knowledge platform.

Engineering Service / Platform Mapping¶

Siemens NX — the deepest relationship: Solution Partner since 2010; Moldex3D eXplorer and SYNC plug-ins; the Siemens-branded NX EasyFill / Easy Fill Advanced runs a Moldex3D subroutine; DesignSim (Jul 2026) is built into NX (versions 2312–2606) with Teamcenter-adjacent workflows. Notably, the DesignSim launch came after Siemens completed its Altair acquisition (2025) — evidence the relationship continues post-Altair; no public statement of change was found either way. [Evidence] ¹⁴¹³¹²
PTC Creo — PTC PartnerAdvantage member; eDesignSYNC → SYNC plug-in with full fill/pack/cool/warp inside Creo. [Evidence] ¹⁵¹⁶
Ansys — listed Ansys technology partner (Workbench-ecosystem interoperability; fiber-orientation mapping to structural analysis). [Evidence — listing] ¹⁷
Dassault Systèmes — listed CoreTech partner page exists; Moldex3D historically offered SYNC for SOLIDWORKS. Depth of the current relationship unverified. [Evidence — listing] ¹⁸
LS-DYNA / Atlas — 2026 release exports fiber-orientation results for downstream structural simulation. [Evidence] ²
CATIA — no native plug-in verified; CATIA users reach Moldex3D via STEP or neutral formats. Relevant to this study's CATPart context. [Evidence of absence — thin]

Moldex3D's structural role: the moulding engine inside everyone else's platform. [Synthesis]

Engineering Intelligence Stack Mapping¶

Intent — weak. No requirements capture; intent enters as simulation objectives and parameter ranges. [Evidence of absence]
Knowledge — the strongest moulding-knowledge layer of any vendor studied. iSLM turns a company's own projects into a searchable, similarity- queryable knowledge base; Material Hub Cloud adds 8,000–9,000+ measured material grades; MoldiBot wraps the vendor's know-how corpus in an LLM. [Evidence] ⁹²⁴⁵
Reasoning — classical solver physics; DoE/optimization search; nascent simulation-trained surrogates (Process Discovery). [Evidence + thin]
Execution — simulation automation (API, Auto Launch, automated meshing); no design authoring. [Evidence] ³
Feedback — closer than anyone else, and still short. iSLM stores mould- trial results next to simulations; iMolding Hub characterizes real machines; but no product learns a model from measured production outcomes (defects, scrap, warpage as-built). The loop is archived, not yet learned from. [Synthesis]

Strengths¶

Pure-play depth: the entire company is injection moulding — solver, materials lab, machine data, knowledge platform. [Evidence] ¹⁹
Ubiquity inside other platforms (NX, Creo, Ansys ecosystem) — it wins even when the customer's platform is someone else's. [Evidence] ¹³¹⁵¹⁷
iSLM Discovery is the first shipping moulding knowledge-retrieval product found in this study — similarity search over historical mould designs with transparent provenance. [Evidence] ⁸
Measured-data moat: ISO 17025 lab + 8,000–9,000+ material grades — the expensive, proprietary part of moulding simulation. [Evidence] ²⁴
Honest labelling in places: "DOE Wizard" is called DoE; DesignSim is described as rule-based. [Evidence] ²¹¹

Weaknesses¶

The flagship "AI" is thinly documented. No architecture, accuracy figure or benchmark exists for the AI Optimization Wizard, Process Discovery or the Discovery similarity matching. [Evidence of absence]
The LLM is bought in (ChatGPT) — same structural dependency as every other vendor. [Evidence] ⁵
No learning from production outcomes — predictions are simulation-taught; the factory's real defect history remains untapped. [Synthesis]
Private and opaque: no financials, no investor scrutiny of AI claims, no open source, small headcount (~200) limiting AI R&D capacity. [Evidence] ²¹³¹
Ecosystem lock: iSLM's knowledge value accrues only inside the Moldex3D tool chain; a CATIA-centred company gets no native plug-in. [Inference]

Current Gaps (largely manual today)¶

Interpreting simulation results still falls to engineers (MoldiBot answers how-to questions; it does not read your warpage plot). Correlating simulation predictions with actual mould-trial outcomes is stored in iSLM but analysed by humans. Gate/runner/cooling design remains expert work with advisory tools. No product predicts defects or scrap from a company's measured production history. [Evidence + Inference]

Future Direction¶

Continue the A.O.I. arc: more Discovery-style retrieval, deeper LLM integration (Moldibot "Deep Thinking", AI Chat), more automation. [Evidence] ²
Push DesignSim-style DFM into CAD seats — expanding from analyst tools to designer tools inside NX (and plausibly Creo next). [Evidence + Inference] ¹²
The announced LLM cooling-channel generation signals ambitions in generative tool design; watch whether it ships. [Evidence — thin] ⁴
My read: CoreTech's strategy is to convert its two real moats — measured material data and customers' accumulated project data — into recommendation products, while renting the LLM layer. The unclaimed prize, learning from production outcomes, would require instrumented customers and is not yet visible in the roadmap. [Inference]

Relevance to Automotive Moulded Parts¶

Capability	Strength	Notes
Plastic Part Design	Partial	Advisory only — mouldability feedback into others' CAD
Surface Design (Class A)	None	Not a CAD vendor
CAD Automation	Partial	API/Python automation of simulation, not of design
Mould Flow	The core product	Fill/pack/cool/warp, fibers, multi-shot, potting; embedded in NX/Creo
Tool Design	Partial	Cooling Channel Designer, gate advisers; LLM cooling-channel generation announced, unshipped
DFM	Strong — rules-based	DesignSim in NX (Jul 2026): undercuts, draft, sink marks, L/t — explicitly rule-based
Manufacturing Engineering	Partial	iMolding Hub machine characterization, trial guidance
Quality	Partial	Simulation-vs-trial records in iSLM; no outcome-learning
Engineering Knowledge Reuse	Strongest of any vendor studied	iSLM Discovery: similarity retrieval over historical mould projects

Critical finding. The study's spine question for this vendor was: is the claimed AI genuinely AI/ML, or relabelled DoE/optimization? The honest answer is both, in separable layers:

Genuine and shipping: an LLM assistant (MoldiBot — real, but generic bought-in technology) and similarity-retrieval over historical moulding projects (iSLM Discovery — real, moulding-specific, and the first of its kind found shipping; retrieval rather than learning).
Genuine but thin: Process Discovery's deep-learning surrogate trained on simulation batches — the first vendor ML claim found that actually targets moulding process prediction; undocumented beyond marketing-level text.
Relabelled: the "AI Optimization Wizard" — on available evidence, automated parameter search in the DoE tradition³².

Net. Moldex3D is the first vendor in this study whose AI portfolio actually touches the moulding domain — but its most substantive shipping capability is knowledge retrieval, not machine learning, and nothing anywhere in the portfolio learns from measured production outcomes. [Synthesis]

Key Takeaways¶

CoreTech System (Taiwan, 1995, ~200 staff) is private, not listed — no public financials exist; the "one of two dominant mould-flow solvers" framing is industry-standard but unaudited. [Evidence] ²¹³¹
Its structural role is the moulding engine inside other vendors' platforms — Siemens NX EasyFill is Moldex3D under the hood; plug-ins for Creo; Ansys and Dassault partner listings. [Evidence] ¹³¹⁵¹⁷
Its AI genuinely reaches the moulding domain — the first vendor studied for which that is true. [Synthesis]
The most substantive shipping capability is iSLM Discovery: similarity retrieval over a company's historical mould projects — retrieval, not learning, and the closest thing to a moulding knowledge base on the market. [Evidence] ⁸
Process Discovery is the first vendor claim of a deep-learning surrogate for moulding process prediction — trained on simulation runs, not factory outcomes, and documented only thinly. [Evidence — thin] ⁴
MoldiBot is a real LLM assistant (ChatGPT + RAG over vendor know-how), announced December 2024 — useful, generic, bought-in. [Evidence] ⁵
The "AI Optimization Wizard" shows no disclosed learning mechanism — treat as automated DoE/optimization under an AI label³². The separately named DOE Wizard and rule-based DesignSim are honestly labelled. [Synthesis]
"A.O.I." = Automation, Optimization, Intelligence — the Moldex3D 2026 release framing, not a product. [Evidence] ²
All quantified automotive customer evidence (DURA, car-roof, Frumolde) predates the AI features — it proves the solver, not the AI. [Evidence] ²⁷
Nothing in the portfolio learns from measured production outcomes — the labeled-data gap survives even at the moulding specialist. [Synthesis]

References¶

Primary Sources — CoreTech / Moldex3D¶

Moldex3D 2025 "Molding Intelligence" release, Mar 26 2025 — https://www.moldex3d.com/news/moldex3d-releases-version-2025/
Moldex3D 2026 "A.O.I." release, Mar 20 2026 — https://www.moldex3d.com/news/moldex3d-releases-version-2026/
Moldex3D 2025 product page (feature detail) — https://www.moldex3d.com/products/moldex3d-2025/
MoldiBot announcement, Dec 16 2024 — https://www.moldex3d.com/news/moldibot-debuts-on-moldiverse-optimizing-customer-support/
iSLM product page — https://www.moldex3d.com/products/moldex3d-islm/
iSLM Discovery innovations blog — https://www.moldex3d.com/blog/top-story/innovations-in-islm-driving-intelligent-data-driven-decision-making/
iSLM big-data platform blog — https://www.moldex3d.com/blog/top-story/realizing-cyber-physical-integration-through-the-big-data-management-platform/
Moldiverse launch blog — https://www.moldex3d.com/blog/top-story/the-brand-new-moldiverse-cloud-platform-takes-the-first-step-towards-industry-transformation/
Material Hub Cloud — https://www.moldex3dmaterial.com/
About Moldex3D — https://www.moldex3d.com/about/ (Chinese: https://ch.moldex3d.com/about/)
Leadership transition, Apr 21 2022 — https://www.moldex3d.com/news/20220421-2/
NX EasyFill integration, Dec 10 2012 — https://www.moldex3d.com/news/20121210-2/
Moldex3D eXplorer for NX release — https://www.moldex3d.com/news/moldex3d-announced-the-official-release-of-nx-plug-in-moldex3d-explorer-for-nx/
eDesignSYNC for Creo release — https://www.moldex3d.com/news/moldex3d-released-a-new-interface-for-creo-elementspro-edesignsync-for-creo/
SYNC 2022 for PTC Creo docs — https://support.moldex3d.com/2022/sync/sync-for-ptc-creo/1-1_introductiontosync.html
Fiber-orientation patent US 8,571,828, Oct 15 2013 — https://www.moldex3d.com/news/20131015-2/
Best of ANTEC 2025 award — https://www.moldex3d.com/news/honored-with-best-of-antec-2025-moldex3d-drives-innovation-in-electronics-potting/
AI + iSLM webinar listing — https://www.moldex3d.com/events/webinar/webinar-ai-islm-turn-moldex3d-simulation-results-into-reusable-design-decision-assets-cet/

Customer Evidence (vendor-published)¶

DURA Automotive window encapsulation, Oct 3 2019 — https://www.moldex3d.com/blog/customer_success/moldex3d-helps-dura-automotive-solve-the-deformation-of-a-vehicle-window-encapsulation/
Car roof cartridge warpage optimization — https://www.moldex3d.com/blog/customer_success/warpage-optimization-of-automotive-car-roof-cartridge-part/
"World's leading automotive part supplier" — https://www.moldex3d.com/blog/customer_success/worlds-leading-automotive-part-supplier-utilizes-moldex3d-analyses-to-make-confident-decisions-on-product-design-optimization/
Frumolde Tooling testimonial — https://www.moldex3d.com/blog/customer-testimonials/frumolde-tooling-a-market-leader-in-automotive-mold-design-and-manufacturing-through-moldex3d-integration/

Partner Listings¶

Ansys technology partner page — https://www.ansys.com/partner-ecosystem/technology-partners/coretech-system-company
Dassault Systèmes partner listing — https://www.3ds.com/partners/partner-details/200000000029922_CORETECH_SYSTEM_CO_LTD

Secondary Sources¶

SPE Injection Molding Division: "Moldex3D 2025 Molding Intelligence" (Tom Wood), Apr 22 2025 — https://www.injectionmoldingdivision.org/2025/04/22/moldex3d-2025-molding-intelligence-embracing-the-ai-revolution-in-injection-molding/
Engineering.com: Moldex3D 2025 release — https://www.engineering.com/coretech-releases-moldex3d-2025-for-molding-intelligence/
Engineering.com: Moldex3D 2026 A.O.I. — https://www.engineering.com/moldex3d-releases-2026-version-with-a-o-i-updates/
Engineering.com: DesignSim announcement — https://www.engineering.com/coretech-announces-designsim-for-dfm-validation-in-siemens-nx/
DailyCADCAM: DesignSim launch, Jul 16 2026 — https://dailycadcam.com/moldex3d-launches-designsim-embedding-injection-molding-dfm-validations-in-nx/
Taiwan unlisted-stock tracker (科盛科技未上市) — https://ipo168.pixnet.net/blog/post/207622297
ZoomInfo company profile (employee count) — https://www.zoominfo.com/c/coretech-system-co-ltd-moldex3d/100516498

Appendix A — Timeline¶

1995 founding (five academics, NTHU lineage) → 2010 Siemens partner program → 2012 NX EasyFill (Moldex3D inside NX) → 2013 fiber-orientation patent → ~2018–20 iSLM data platform → 2022 leadership transition (Hsu Chairman, Yang CEO) → 2023 Moldiverse cloud → Dec 2024 MoldiBot (ChatGPT + RAG) → Mar 2025 Moldex3D 2025 "Molding Intelligence" (AI Optimization Wizard, iSLM Mold/Gate Design Discovery, Process Discovery, LLM cooling-channel tech announced) → Apr 2025 Best of ANTEC award → Mar 2026 Moldex3D 2026 "A.O.I." (Moldibot Deep Thinking, AI Chat, AI Material Assistant, Gate Location Advisor, DOE Wizard constraints) → Jul 2026 DesignSim (rule-based DFM in NX).

Appendix B — Glossary¶

Mould-flow (CAE) simulation — numerically solving how molten plastic fills, packs, cools and warps in a mould; Moldex3D's core business.
A.O.I. — "Automation, Optimization, Intelligence": the Moldex3D 2026 release framing (not a product, and not the electronics-industry meaning "automated optical inspection").
iSLM — Moldex3D's data-management/knowledge platform for moulding projects. The vendor's own materials expand the name inconsistently ("intelligent Simulation Lifecycle Management" / "Intelligent Solution Lifecycle Management"); this report uses the abbreviation only.
Discovery (iSLM) — the 2025–26 similarity-retrieval features: match a new part's geometry against historical projects and surface their gate/mould designs and outcomes as predictions.
Process Discovery — claimed deep-learning model trained on a project's simulation runs to predict moulding results in real time during machine setup.
Surrogate model — a machine-learning model trained on simulation results that then predicts outcomes near-instantly without re-running the physics.
DoE (design of experiments) — running a structured grid of simulations across parameter ranges and picking the best performer; automated search, not learning.
RAG (retrieval-augmented generation) — an LLM answering from documents retrieved at question time; MoldiBot's mechanism.
Moldiverse — CoreTech's cloud ecosystem (materials, machine data, training, MoldiBot).
Material Hub Cloud (MHC) — online database of 8,000–9,000+ lab-measured plastic grades.
DesignSim — rule-based DFM validation tool inside Siemens NX (Jul 2026).
DFM — design for manufacturability: checking a design against the rules that make it mouldable (draft angles, wall thickness, undercuts…).

Appendix C — Notes (thinnest-evidence areas to revisit)¶

Process Discovery's "deep learning" — the single most important thin claim: no architecture, accuracy data or independent validation anywhere; sourced to vendor text relayed by an SPE division article. If substantiated, it upgrades the report's central finding; if not, the "real ML" layer shrinks to the LLM chatbot. Re-checked 2026-08-02: still no architecture or accuracy disclosure anywhere. The Moldex3D 2026 release news (Mar 20, 2026) does not mention Process Discovery by name (emphasis moved to the iSLM Discovery series, AI Chat and Moldibot); an Apr 15, 2026 "AI + iSLM" webinar describes the retrieval Discovery series (Insight/Design Discovery, Auto Launch) but again publishes no numbers. The workflow chain (Molding Window Advisor DoE → Process Discovery surrogate → iMolding Hub deployment) is confirmed; the absence of any public accuracy figure remains a first-class negative finding. Item stays OPEN.
iSLM "AI shape recognition" — similarity mechanism (learned embeddings vs. geometric descriptors) undisclosed.
AI Optimization Wizard mechanism — classified here as relabelled optimization by absence of any disclosed learning mechanism; CoreTech has never published how it works.
LLM cooling-channel generation — announced technology; no product page, demo or shipping evidence found.
AI Material Assistant (2026) — one-line description only.
Employee count (~192) and "world's second-largest" market position — both secondary-source; no audited figures exist (private company).
Patent US 11,220,033 — plausibly CoreTech's, assignee unverified; do not cite as theirs.
RESOLVED 2026-08-02: the "world's leading automotive part supplier" case is confirmed as Faurecia Interior Systems India Pvt. Ltd. (Pune) — the case page itself names the customer, verified by direct fetch during the Forvia report research and recorded in that report's Public Customer Evidence section (see the Forvia report). §7 updated accordingly.
Siemens relationship post-Altair — DesignSim's July 2026 NX launch implies continuity, but no explicit statement from either side was found.
iSLM full product name and launch year — vendor materials are inconsistent; the ~2018–2020 dating here is inferred from blog archaeology.

Executive Summary¶

Who they are. CoreTech System is a Taiwanese company founded in 1995 by five academics. Its single product line, Moldex3D, is one of the two dominant independent injection-moulding simulation ("mould-flow") solvers, alongside Autodesk Moldflow. It is privately held (not listed on any Taiwanese exchange — a correction to common assumptions), with roughly 200 employees and offices across Asia, Europe and North America. [Evidence] ¹⁹²¹³¹
Why they matter for moulded parts. Moldex3D is a pure-play: injection moulding is not one module among many — it is the entire company. Its solver is also embedded inside other vendors' platforms (Siemens NX "EasyFill Analysis" is powered by Moldex3D; SYNC plug-ins exist for NX and PTC Creo; it is a listed Ansys and Dassault technology partner). If genuine AI-for-moulding were going to appear anywhere first, it would be here. [Evidence] ¹³¹⁵¹⁷¹⁸
AI maturity: the most moulding-specific AI of any vendor studied — but mixed. Under the 2025 "Molding Intelligence" and 2026 "A.O.I." (Automation, Optimization, Intelligence) banners, Moldex3D shipped: a ChatGPT-based support assistant (MoldiBot, announced December 2024); a similarity-retrieval knowledge base over a company's own historical moulding projects (iSLM Discovery); a claimed deep-learning process-prediction model (Process Discovery); and an announced LLM-driven cooling-channel generator. Alongside these sit tools whose "AI" label covers classical search: the AI Optimization Wizard (mechanism undisclosed; behaves like automated design-of-experiments) and an explicitly named DOE Wizard. [Evidence] ¹²⁴⁵
The central honest finding. Moldex3D is the first vendor in this study whose AI genuinely reaches the injection-moulding domain — but not in the way the marketing suggests. The genuinely novel, shipping capability is knowledge retrieval (iSLM's similarity search over past mould designs), plus a generic LLM chatbot. The claimed machine learning for process prediction (Process Discovery) appears real but is trained on the company's own simulation runs, not on measured factory outcomes, and is thinly documented. The flagship "AI Optimization Wizard" is, on the available evidence, automated parameter search wearing an AI label. [Synthesis] ⁴⁸

Moldex3D 2025 "Molding Intelligence" release, Mar 26 2025 — https://www.moldex3d.com/news/moldex3d-releases-version-2025/ ↩↩↩↩↩↩
Moldex3D 2026 "A.O.I." release (Automation, Optimization, Intelligence pillars), Mar 20 2026 — https://www.moldex3d.com/news/moldex3d-releases-version-2026/ ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
Moldex3D 2025 product page (AI Optimization Wizard, Molding Window Advisor, iSLM Auto Launch, API) — https://www.moldex3d.com/products/moldex3d-2025/ ↩↩↩↩↩↩
SPE Injection Molding Division, "Moldex3D 2025 Molding Intelligence" (Tom Wood), Apr 22 2025 — mechanisms: ChatGPT + RAG for MoldiBot; "deep learning data science methods" for Process Discovery; LLM-prompted cooling channels — https://www.injectionmoldingdivision.org/2025/04/22/moldex3d-2025-molding-intelligence-embracing-the-ai-revolution-in-injection-molding/ ↩↩↩↩↩↩↩↩↩↩↩↩↩↩
"Intelligent Assistant MoldiBot Soon Makes Its Debut," Dec 16 2024 — "built on ChatGPT technology" — https://www.moldex3d.com/news/moldibot-debuts-on-moldiverse-optimizing-customer-support/ ↩↩↩↩↩↩↩↩
Engineering.com, "CoreTech releases Moldex3D 2025 for molding intelligence" — https://www.engineering.com/coretech-releases-moldex3d-2025-for-molding-intelligence/ ↩
Engineering.com, "Moldex3D releases 2026 version with A.O.I. updates" — https://www.engineering.com/moldex3d-releases-2026-version-with-a-o-i-updates/ ↩
"Innovations in iSLM Driving Intelligent Data-driven Decision-making" — similarity comparisons against historical projects; provenance transparency — https://www.moldex3d.com/blog/top-story/innovations-in-islm-driving-intelligent-data-driven-decision-making/ ↩↩↩↩↩↩
iSLM product page — https://www.moldex3d.com/products/moldex3d-islm/ ↩↩↩↩
iSLM big-data management platform blog (Knowledge Management search) — https://www.moldex3d.com/blog/top-story/realizing-cyber-physical-integration-through-the-big-data-management-platform/ ↩
DailyCADCAM, "Moldex3D Launches DesignSim, Embedding Injection Molding DFM Validations in NX," Jul 16 2026 — "rule-based DFM analysis" — https://dailycadcam.com/moldex3d-launches-designsim-embedding-injection-molding-dfm-validations-in-nx/ ↩↩↩↩
Engineering.com, "CoreTech announces DesignSim for DFM validation in Siemens NX" — https://www.engineering.com/coretech-announces-designsim-for-dfm-validation-in-siemens-nx/ ↩↩↩↩↩
Moldex3D filling simulation integrated with Siemens NX 8.5 (EasyFill Analysis), Dec 10 2012 — https://www.moldex3d.com/news/20121210-2/ ↩↩↩↩↩↩
Moldex3D eXplorer for NX release (Siemens partner program since Jan 2010) — https://www.moldex3d.com/news/moldex3d-announced-the-official-release-of-nx-plug-in-moldex3d-explorer-for-nx/ ↩↩↩
eDesignSYNC for Creo Elements/Pro release (PTC PartnerAdvantage) — https://www.moldex3d.com/news/moldex3d-released-a-new-interface-for-creo-elementspro-edesignsync-for-creo/ ↩↩↩↩↩
Moldex3D SYNC 2022 for PTC Creo documentation — https://support.moldex3d.com/2022/sync/sync-for-ptc-creo/1-1_introductiontosync.html ↩↩
CoreTech System on Ansys technology-partner directory — https://www.ansys.com/partner-ecosystem/technology-partners/coretech-system-company ↩↩↩↩
CoreTech System partner listing at Dassault Systèmes — https://www.3ds.com/partners/partner-details/200000000029922_CORETECH_SYSTEM_CO_LTD ↩↩
About Moldex3D (founding, founders, positioning) — https://www.moldex3d.com/about/ ↩↩↩
Moldex3D Chinese about page / Chinese business press ("Asia's largest, world's second-largest" framing) — https://ch.moldex3d.com/about/ ↩↩
Taiwan unlisted-stock tracker: 科盛科技 is 未上市 (not listed on main/OTC/emerging boards); founded Nov 1995 — https://ipo168.pixnet.net/blog/post/207622297 ↩↩↩↩↩↩
"A Generation Transition: CoreTech System Appointed New Chairman and CEO," Apr 21 2022 — https://www.moldex3d.com/news/20220421-2/ ↩
"The Brand New Moldiverse Cloud Platform…" (2023 launch) — https://www.moldex3d.com/blog/top-story/the-brand-new-moldiverse-cloud-platform-takes-the-first-step-towards-industry-transformation/ ↩↩↩↩
Moldex3D Material Hub Cloud (8,000–9,000+ grades; ISO 17025 lab) — https://www.moldex3dmaterial.com/ ↩↩↩↩↩↩
US patent 8,571,828 fiber-orientation model (iARD-RPR), Oct 15 2013 — https://www.moldex3d.com/news/20131015-2/ ↩↩
"Best of ANTEC 2025" award with Chao Long Motor Parts (electronics potting) — https://www.moldex3d.com/news/honored-with-best-of-antec-2025-moldex3d-drives-innovation-in-electronics-potting/ ↩↩↩↩
DURA Automotive window-encapsulation case study, Oct 3 2019 — https://www.moldex3d.com/blog/customer_success/moldex3d-helps-dura-automotive-solve-the-deformation-of-a-vehicle-window-encapsulation/ ↩↩↩
Car roof cartridge warpage-optimization case study — https://www.moldex3d.com/blog/customer_success/warpage-optimization-of-automotive-car-roof-cartridge-part/ ↩↩
"World's Leading Automotive Part Supplier Utilizes Moldex3D Analyses…" (customer unnamed in title) — https://www.moldex3d.com/blog/customer_success/worlds-leading-automotive-part-supplier-utilizes-moldex3d-analyses-to-make-confident-decisions-on-product-design-optimization/ ↩
Frumolde Tooling customer testimonial — https://www.moldex3d.com/blog/customer-testimonials/frumolde-tooling-a-market-leader-in-automotive-mold-design-and-manufacturing-through-moldex3d-integration/ ↩↩
ZoomInfo profile, CoreTech System Co., Ltd. (~192 employees; secondary source) — https://www.zoominfo.com/c/coretech-system-co-ltd-moldex3d/100516498 ↩↩↩↩
What "optimization technology" means and why it is not generative AI — see the concept note What is Optimization technology: automated parameter search / virtual DoE runs simulations across candidate settings and picks the best performer; it learns nothing from data, so it is classical optimization, not AI. ↩↩↩↩↩↩
Frumolde — Marinha Grande, Portugal — https://www.frumolde.com ↩