Japan’s Gear, Taiwan’s Fabs: How Tricuss’ AI Accelerates Semiconductor R&D for Both

This startup has been selected by StarFab Accelerator based in Hsinchu, Taiwan, and is scheduled to exhibit at The 12th Cloud Computing Day Tokyo on October 31, 2025.

The year 2025 is being hailed as the “Year of AI Agents” worldwide. From consumer-facing AI assistants to sophisticated enterprise decision-support systems, AI technology is evolving rapidly. Among these developments, Taiwan-based Tricuss has carved out a remarkably unique position. The company’s “Co-Researcher AI Agents” offer an innovative solution that dramatically enhances the capabilities of researchers and engineers in semiconductor manufacturing—one of the world’s most complex and data-intensive industries.

The company name “Tricuss” derives from the English word “Discuss.” While “Discuss” traditionally means “a dialogue between two people,” adding the prefix “Tri” (three) expresses the company’s vision that “AI will participate in future discussions.” The name embodies their goal of transcending simple two-way dialogue to create a new form of collaboration where humans and AI work together. Founder and CEO Nai-Hsiang Wang (pictured above) has an unconventional background with a graduate degree in architecture and studies in both engineering and philosophy—a combination that brings a unique perspective to the company’s product design.

Since its founding in May 2022, Tricuss has built a track record with approximately 20 mid-sized and large enterprises. Customers include leading Taiwanese semiconductor and electronics component manufacturers such as AUO, MiTAC, and Hermes Epitek. As a top partner in Nvidia’s Inception Program, with speaking engagements at Nvidia GTC and a win at Intel’s Edge Solutions Challenge, the company has earned exceptionally high technical recognition. Now, as the company prepares for a pre-seed funding round in 2025, it’s setting its sights on full-scale expansion into the Japanese market.

The $10,000 Daily Cost and the Statistics Wall

In semiconductor manufacturing, optimizing “recipes” (manufacturing process parameters) is a critically important factor that determines product quality and profitability. In reality, however, this optimization process still relies heavily on trial-and-error approaches. Researchers must analyze over 500 sensors and parameters—a level of complexity incomparable to other manufacturing industries.

Wang explains the challenges facing the industry with concrete figures.

Even in small-scale experimental facilities, a single wafer experiment costs about $10,000. And since finding the optimal recipe requires 10 to 12 experiments. We’re looking at $10,000 in expenses every day. (Wang)

Even more serious is the time pressure. In the highly competitive semiconductor industry, top companies like TSMC need to find the right recipe as quickly as possible. Research shows that a six-month delay in product launch can reduce a high-tech company’s profitability by as much as 33%. These time and cost constraints place enormous pressure on the entire industry.

However, the most fundamental problem is the skills gap in talent. In semiconductor manufacturing, statistical methods are indispensable. Researchers need to master advanced analytical techniques such as Design of Experiments (DOE), Taguchi methods, multivariate analysis, and even machine learning models. In reality, though, many researchers are experts in chemistry or materials science but don’t necessarily have deep knowledge of statistics.

Data in the semiconductor industry is orders of magnitude more complex than in other manufacturing sectors. Many researchers can only do simple analysis in Excel, and personnel who can skillfully use statistical tools like Minitab or SPSS are limited. (Wang)

This situation creates a contradiction: companies invest heavily in digitalization, install massive numbers of IoT sensors to collect data, yet cannot actually utilize that data for decision-making. The data exists, but there’s a shortage of personnel who can analyze it and transform it into actionable insights. This massive gap in the transformation to “data-driven enterprises” is precisely the problem Tricuss is trying to solve.

A Virtual Statistics Expert at Your Side

What Tricuss offers is not simply a data analysis tool. The company’s platform provides all researchers and engineers with an experience akin to having a statistics PhD working alongside them at all times.

The “Data Researcher AI Agent” at the system’s core is fully integrated with a company’s existing workflows, systems, and data (both structured and unstructured). It supports over 400 types of data sources, from standard protocols like REST API, MQTT, and SQL Server, to enterprise systems like Oracle, PostgreSQL, MySQL, SAP HANA, and Salesforce, and even IoT devices.

Our core researcher engine is like having a statistics PhD working with you. (Wang)

Users can combine their own domain expertise (such as knowledge of chemical materials) with the AI agent’s statistical expertise. The result is comprehensive analysis that includes not just quantitative data but qualitative insights as well.

The value delivered by the company’s solution can be expressed in four dimensions:

• Depth — From surface-level analysis to root cause identification, prediction, and optimization. The system automatically applies over 10 statistical methods and machine learning models (regression analysis, anomaly detection, predictive maintenance, supply chain optimization, etc.) to identify which parameters (temperature, pressure, flow rate, etc.) are causing problems. Analysis that previously required master’s-level knowledge in statistics or AI can now be performed by anyone.
• Speed — In actual deployment cases, the system analyzes 10 million data points in 12 seconds. Everything from database connection, SQL query creation, analysis execution, chart drawing, and insight delivery is completed within 12 seconds. Work that previously took days is now completed in mere seconds.
• Reach — Not just experts, but all researchers and engineers can access advanced statistical analysis. This improves data literacy across the organization, enabling more people to make evidence-based decisions.
• Proactivity — Enables a shift from reactive to predictive decision-making. Rather than responding after problems occur, the AI agent predicts issues in advance and recommends preventive actions.

Previously, researchers needed to master complex statistical tools like Minitab and SPSS, but Tricuss’s agent incorporates the expertise of these tools. Users simply ask questions in natural language, and the AI agent selects the appropriate statistical methods, executes the analysis, and interprets the results. This represents a democratization of data analysis.

What sets Tricuss’s system apart from other analysis tools is that it doesn’t just provide numbers and graphs—it combines document understanding capabilities to present actionable recommendations.

For example, suppose a researcher sets a goal to “keep uniformity below 3%.” The AI agent provides comprehensive support as follows:

First, it uses appropriate statistical methods (such as ANOVA, correlation analysis, principal component analysis, etc.) to analyze the data and specifically indicates which parameters (flow rate, temperature, pressure, etc.) should be adjusted and how. But Tricuss’s true innovation starts here.

Based on metrics obtained from the analysis (e.g., deviation of a specific factor exceeds 3%), the system automatically searches the company’s internal documents. This includes manuals, SOPs (Standard Operating Procedures), past research reports, technical papers—all document formats. By combining OCR (Optical Character Recognition) with LLMs (Large Language Models), it can process PDFs, Word files, Excel spreadsheets, PowerPoint presentations, and even image files.

In actual use cases, the AI agent provides recommendations such as:

• Yield Improvement — For high-frequency defects (e.g., bright state defects), it recommends checking equipment production parameters and improving process tension to reduce defect occurrence.
• Equipment Inspection — For non-periodic defects, it instructs checking the relative position of rollers, and after taking action, conducting further AOI (Automated Optical Inspection) to confirm improvements.
• Product Risk Management — Even when QC (Quality Control) findings detect no anomalies, the suggested handling method is to hold items with main defect DT values above 0.5, have quality assurance personnel review them, and determine next steps.

Furthermore, the system converts company research papers and technical documents into graph databases and vector databases. This enables the generation of more comprehensive recommendations that combine quantitative analysis results with qualitative information such as past research findings and expert opinions. For example, when a specific parameter adjustment is determined necessary, information about what results were obtained in past similar cases and what findings exist in related academic papers is also provided.

One of Tricuss’s technical strengths is addressing the LLM “hallucination” problem. The company employs hybrid search algorithms to achieve accurate citation and referencing of information. While many other vendors promise 80% accuracy, Tricuss guarantees 95% accuracy, with some actual cases achieving 99% accuracy. This is an extremely important element for enterprise decision-support tools.

Simulation Integration and Complete On-Premises Deployment

Tricuss’s innovation extends not only to physical experiments but also to integration with simulation tools and enterprise architecture that operates completely on-premises.

The AI agent integrates with simulation software and functions as a fully autonomous researcher. The specific process consists of the following cycle:

First, the AI agent uses machine learning models to propose parameter sets considered optimal. Next, through an MCP (Message Communication Protocol) server or automation scripts, it automatically inputs new parameters into the simulation tool and initiates simulation experiments. When the simulation completes, the AI agent retrieves and evaluates the results.

If results don’t meet target values, the AI agent uses machine learning models to improve the analytical model and proposes new parameter sets. This cycle continues autonomously until an optimal design or recipe is found.

Even more noteworthy is Tricuss’s proprietary search algorithm. Compared to traditional Design of Experiments (DOE) methods, it can achieve efficiency improvements of up to 10 million times in 10-dimensional parameter space, dramatically reducing the number of experiments needed to reach optimal solutions.

The company is also collaborating with Nvidia’s Omniverse solution. Omniverse is an integrated platform for physical simulation, and when combined with Tricuss’s data analysis and optimization system, it can be utilized in digital twin and virtual factory scenarios.

Nvidia wants to expand its Omniverse into simulation use cases, and we’re advancing integration with our data analysis system. (Wang)

The advantages of this approach are multifaceted. First, physical experiment costs ($10,000 per run) can be significantly reduced. Second, experiment parallelization becomes easier, allowing simultaneous testing of multiple parameter sets. Furthermore, the ability to conduct bold experiments without fear of failure increases the possibility of discovering more innovative solutions.

In the semiconductor industry, manufacturing recipes and process data are top-level trade secrets. Tricuss deeply understands this and has designed a system that operates completely on-premises.

We are not an OpenAI wrapper. We train our own models. (Wang)

The company fine-tunes models based on open-source language models (such as Llama, Mistral, Qwen, etc.) for enterprise-specific needs, including industrial algorithms, statistical methods, optimization techniques, simulation technology, and machine learning/deep learning training scripts. All algorithms operate on-premises, with no internet connection required. This is to meet the strict data security requirements demanded by top semiconductor companies like TSMC.

The system architecture satisfies four elements necessary for enterprise-grade AI agents:

• Integration — Integration with existing enterprise workflows, systems, and structured/unstructured data. Broad integration is possible from core systems like ERP, CRM, and MES to various document management systems.
• Data Scale — Handles large-scale data at the TB (terabyte) to PB (petabyte) level and decades of historical documents. Even at a single department level, it can handle billions of records.
• Complex Scenarios — Handles not simple Q&A but complex processes of data analysis and decision support. Can execute advanced cognitive tasks such as multi-stage reasoning, integration of multiple data sources, and causal analysis.
• Security — Data isolation, on-premises deployment, hybrid cloud support. All data processing is completed within the company’s firewall, minimizing the risk of external information leakage.

According to Wang, many companies have tried building their own AI systems or simply tried other providers using GPT APIs, but “delivery took twice as long and failed to meet enterprise standards.”

Tricuss is positioned as bridging the “enormous gap” between consumer AI and enterprise-grade AI. While consumer-facing ChatGPT is certainly convenient, enterprise systems that handle confidential corporate data, analyze with 99%+ accuracy, integrate with existing systems, and operate stably 24/7/365 have completely different requirements.

The company also provides complete GPU server solutions and, as a certified Nvidia partner, can offer integrated solutions from hardware to software. This enables customers to deploy AI agent systems in a one-stop manner without needing to deal with multiple vendors.

Strategic Japan Market Expansion

Tricuss has clear strategic reasons for prioritizing the Japanese market. Wang highly values the strengths of Japan’s semiconductor industry.

In semiconductor manufacturing equipment, Japan remains a leader with over 60% share of the global market. They’re also top-class in materials. (Wang)

Indeed, equipment manufacturers like Tokyo Electron, Screen Holdings, and Kokusai Electric, and materials manufacturers like JSR, Shin-Etsu Chemical, and Sumitomo Chemical are indispensable to the global semiconductor supply chain. Meanwhile, in the foundry (contract manufacturing) sector, Taiwan’s TSMC dominates with overwhelming strength.

This complementary relationship is at the core of Tricuss’s strategy. Taiwan, Japan, South Korea, and the United States form the core of the semiconductor value chain, and conquering these markets becomes the foundation for global expansion. However, the strategy for entering the Japanese market is cautious.

Understanding Japanese business culture in a short period is very difficult. Understanding how Japanese companies start business, what the appropriate approaches are—these things take time. (Wang)

Therefore, the company plans an indirect approach through business partners rather than direct sales. Specifically, they’re considering cooperation with system integrators (SI-ers) and value-added resellers (VARs). Taiwanese SI-ers with 20-30 years of business relationships with Japanese companies play particularly important roles as cultural bridges.

We’ll enter the Japanese market through experienced partner companies. They already understand Japanese business practices and have long-standing trust relationships. (Wang)

This strategy is a realistic approach to overcoming cultural barriers and leveraging existing trust relationships. In the semiconductor industry, trust is extremely important, and introductions through trusted partners are more effective than newcomers directly approaching major companies.

For the long-term vision of global expansion, the plan is to first conquer the core semiconductor value chain markets of Taiwan, Japan, South Korea, and the U.S. Subsequently, they’ll expand to European countries (France, Germany, etc.) where factory construction is progressing under the Biden administration’s CHIPS Act.

Due to the Biden administration’s CHIPS Act, countries around the world are trying to build semiconductor factories in their own countries. France and Germany also want to have their own factories. This is an opportunity for us to expand globally. (Wang)

This strategy accurately captures geopolitical trends. As semiconductor supply chains become multipolar, each country is strengthening manufacturing capabilities, and demand for solutions like Tricuss’s is rising accordingly.

For specific target customers in the Japanese market, Wang envisions semiconductor manufacturing equipment makers and materials manufacturers. For such companies, Tricuss’s solution could potentially be offered as a value-added service to their customers (semiconductor manufacturers).

Technical recognition is also extremely high. Tricuss has the following achievements:

• As a top partner in Nvidia’s Inception Program, with speaking engagements at Nvidia GTC. In actual presentations, they demonstrated analyzing 10 million data points in 10 minutes, leaving a strong impression on the audience.
• Won Intel’s Edge Solutions Challenge. Wang emphasizes, “This wasn’t a startup competition but a technical competition where world-class companies competed.” This victory proves that Tricuss’s technical capabilities are not merely startup-level but competitive with global corporations.
• Also participated in Qualcomm’s Innovate in Taiwan Challenge 2025, where their technological innovation in AI was recognized. Relationships with U.S. and global accelerator programs such as Qualcomm, Intel, Play, and Garage+ form an important foundation supporting the company’s international expansion.
• In Taiwan, they’ve also participated in accelerators including Draper University, Microsoft for Startups, and AWS Startups, securing access to cloud infrastructure and startup ecosystems. Additionally, as a fifth-period team at Taiwan Startup Hub, they receive support from the Taiwanese government.

These partnerships are not merely nominal. They’re engaged in technical cooperation with Nvidia on GPU server optimization and Omniverse integration. They’re discussing AI agent deployment in edge computing environments with Intel. Such concrete technical collaborations are accelerating Tricuss’s product development.

Tricuss is planning a pre-seed funding round in 2025. Since its establishment in May 2022, the company has operated on the founder’s own capital and revenue from initial customer projects, but has decided to raise external funding with full-scale expansion into Japan and the U.S. in view.

Currently, they’re in negotiations with venture capitals and strategic investors in both Taiwan and the U.S. While they’ve already received investment from Taiwan’s MIT (Management Innovation Technology), they plan to accelerate product development, sales structure strengthening, and international expansion through further fundraising.

Among Japanese semiconductor-related companies, there are those with corporate venture capital (CVC). Investment from such companies leads not only to funding but also collaboration opportunities. (Wang)

Indeed, when Wang visited the U.S., he realized that many companies use investment as a means of incorporating innovation. Rather than mere financial investment, it’s an approach of investing in startups as part of strategic partnerships and incorporating their technology into one’s own ecosystem.

Tricuss’s business model is a hybrid of enterprise SaaS (Software as a Service) and on-premises licensing. They offer both cloud and on-premises versions according to customer scale and security requirements.

The pricing structure is basically a subscription model based on number of users, but customized contracts are possible for large-scale deployments. They also provide professional services such as consulting during initial implementation, customization, and training.

The company’s revenue structure consists of approximately 70% software license revenue and 30% professional services. As the customer base expands and the product matures, they plan to further increase the proportion of software licenses.

For use of raised funds, they envision product development (40%), sales and marketing (30%), talent recruitment (20%), and operating capital (10%). Wang explains that entering the Japanese market in particular requires establishing local presence, cultural adaptation, and long-term relationship building, which requires appropriate resources.

Accelerating Civilization’s Progress with AI

Wang ‘s background differs from typical tech startup founders. He graduated from an architecture program and studied both engineering and philosophy as an undergraduate. This diverse background brings a unique perspective to Tricuss’s product design.

I wanted to start a business since high school. I participated in various entrepreneurship-related courses and activities and loved the process of building something from scratch. (Wang)

Before founding the company, Wang spent about two years exploring the optimal industry where AI agents could become market leaders. This exploration process demonstrates the company’s strategic caution.

I’m not from the semiconductor industry. However, I worked with TSMC consultants and deeply understood the industry’s challenges. As a result, I found that the AI agent idea of a ‘Co-Researcher in Statistics’ perfectly matched specific pain points in the semiconductor industry. (Wang)

These “pain points” aren’t mere technical challenges but problems with economic impact. Daily costs of $10,000, a 33% profitability decline from a 6-month delay—these are challenges worth significant investment to solve.

What’s important is whether customers are willing to pay sufficient money. Even if technically excellent, if customers don’t recognize value and allocate budget, it won’t become a business. (Wang)

This philosophy is reflected in the company’s product design. Tricuss’s AI agent is designed to be treated not as a mere tool but as a team member. Users dialogue with the AI agent in natural language, ask questions, discuss, and collaboratively solve problems.

Wang also speaks about the conditions for successful founders:

What’s important is decision-making ability. Achieving maximum effect with minimum cost. Judging potential benefits, expected value, risk, and cost-performance, and focusing on what’s most efficient. After founding, this thought process has greatly advanced. (Wang)

Indeed, Tricuss’s strategy is highly calculated. Choosing the high-value semiconductor market, first building a track record in Taiwan, then expanding to key markets of Japan and the U.S., and finally advancing to emerging markets like Europe. Selecting appropriate partners at each stage and fundraising at appropriate timing. This strategic caution forms the foundation of the company’s success.

His management philosophy also shows an attitude that values process over outcome of entrepreneurship. He states that “after founding a company, whether it succeeds or fails, you can grow through experiencing that process,” noting that he’s become able to intuitively judge cost-effectiveness when making decisions, plan and evaluate ideas, and make decisions with a bird’s-eye view perspective.

The Turning Point of the AI Agent Market

The year 2025 is called the “Year of AI Agents,” with rapidly growing interest in AI agent technology worldwide. However, there’s a significant difference between consumer-facing AI assistants and enterprise-grade AI agents like those provided by Tricuss.

The Japanese market presents a compelling opportunity for Tricuss. Facing serious labor shortages due to declining birthrates and aging populations, Japan is actively investing in smart factories, IoT, and digital twins. However, converting the massive data obtained from these technologies into actual value remains a challenge.

Japan’s semiconductor industry is showing signs of revival. With Rapidus’s advanced semiconductor manufacturing project, TSMC’s Kumamoto factory construction, and various government support measures, new factories designed for digitalization and AI utilization from the start present an excellent opportunity to introduce solutions like Tricuss’s.

For future development, the company is also considering expanding applications from current semiconductor manufacturing to smart factories in general.

We already have track records in other manufacturing industries like steel, electronic components, and chemicals. The core technology is the same. Analyzing sensor data, finding optimal parameters, identifying root causes, and providing actionable recommendations. This is applicable to all manufacturing industries. (Wang)

Furthermore, they’re focusing on simulation and optimization in virtual environments, including integration with digital twin technology and Nvidia’s Omniverse. By combining physical experiments, simulation, and AI-driven optimization, they can dramatically improve R&D speed and cost efficiency.

Startups like Tricuss play important roles in Taiwan’s national strategy to fuse AI technology with manufacturing—a vision that aims to maintain the country’s industrial competitiveness for decades to come.

At the interview’s conclusion, Wang reiterated the company’s mission:

Our wish is to enable the world’s most talented people to utilize AI. AI shouldn’t just accomplish tasks but should support human experts in R&D, pharmaceuticals, and other specialized fields to do better work.

As engineers, as craftsmen, we believe we’re working on the highest use case of accelerating civilization’s progress. We may not be able to cure diseases, but we can support the most talented doctors and scientists in creating better simulations, shaping a better future. (Wang)

This philosophy represents a social mission beyond mere profit pursuit. AI technology doesn’t replace human capabilities but augments and empowers them. AI supports routine analytical work so experts can focus on more creative and strategic work. This is the future vision Tricuss aims for.

A Startup Leads Semiconductor Industry Transformation

Tricuss’s challenge represents the cutting edge of AI agent technology. In the highly specialized domain of semiconductor manufacturing, where data is massive and complex and the cost of failure is extremely high, the company’s solution is not merely an efficiency tool but holds the potential to fundamentally transform the entire industry’s R&D process.

Taiwan’s semiconductor industry has built global competitiveness over the past several decades. The foundry business represented by TSMC, IC design companies like MediaTek and Realtek, and numerous equipment and materials suppliers—the ecosystem formed by these companies constitutes the core of the global semiconductor supply chain.

Tricuss represents the next evolution of this ecosystem. In addition to physical manufacturing capabilities and hardware design capabilities, AI-driven intellectual productivity improvements will further strengthen Taiwan’s semiconductor industry’s competitive advantages. Analyzing 10 million data points in 12 seconds, identifying root causes, and providing actionable recommendations—this holds the potential to accelerate R&D speed 100-fold from conventional approaches.

For Japan’s semiconductor industry as well, partners like Tricuss would hold great significance. Japanese companies leading the world in equipment and materials could further strengthen their global competitiveness by adopting cutting-edge AI agent technology.

Starting with the Tokyo visit at the end of 2025, how collaboration between Tricuss and Japanese companies unfolds will be watched with great interest. In the semiconductor industry’s next 10 years, not just physical manufacturing capabilities but how to increase intellectual productivity will become the focus of competition. At the forefront of this, Tricuss holds the potential to play an important role.

Borrowing Wang’s words, “This isn’t the end but the beginning.” Using success in semiconductor manufacturing as a foothold, they’ll expand to smart factories in general, and further to pharmaceuticals, finance, and other data-intensive industries. AI agents will collaborate with human experts and accelerate civilization’s progress. This is the future vision Tricuss envisions.