In line with the widespread dissemination of Internet of Things (IoT) devices and various sensors in fields ranging from industrial settings to people’s daily lives, the volume of data has been increasing rapidly. As a consequence, the effective utilization of such data increasingly affects corporate competitiveness. Highly advanced artificial intelligence (AI) technologies capable of transforming the potential of data into practical value are becoming essential as a powerful engine for achieving sustainable corporate growth as well as social stability. In this context, increasing attention is being focused on trustworthy AI not only for maximizing the use of data but also for implementing risk management.

Toshiba Corporation is committed to innovation and social stability by unlocking the potential of data while advancing both proactive data utilization and defensive risk management through the use of the latest technologies and trustworthy AI.

YAMAGAMI Yuta / FURUTA Tetsuro

Given the frequent occurrence of localized torrential downpours, there is a growing need to ensure the safe and stable operation of social infrastructure, including railways, taking such rainfall into consideration. However, accurately assessing the rainfall conditions necessary for operational decisions is difficult due to the high spatial variability of localized rainfall.

Toshiba Corporation has been providing a variety of weather data services based on its accumulated experience in the development of weather radar data analysis technologies. As part of these efforts, we have been developing a real-time rainfall estimation service capable of monitoring rainfall conditions using highly accurate rainfall estimation technologies. We conducted evaluation experiments on the technologies for this service using an actual railway line and verified that they can observe localized torrential downpours, which have been difficult to predict until now, with a high degree of accuracy and are expected to contribute to enhanced safety and stable railway operations.

WAKE Masahide / YOSHITANI Naoto

Along with the aging of social infrastructure systems and a lack of experienced personnel, there is an urgent need to improve the efficiency of maintenance and inspection and to standardize the quality of condition ratings using artificial intelligence (AI). Increasing attention has been focused on insourcing the development of AI models based on data accumulated by infrastructure operators, as well as on building a common platform to continuously support updates to AI models.

In cooperation with Central Nippon Expressway Company Limited, Toshiba Corporation conducted proof-of-concept (POC) experiments using the SATLYS AI Common Platform and a model-based image anomaly detection AI, and confirmed that personnel without AI expertise can build AI models for detecting pavement distress and rapidly introduce such AI models into actual operations.

KOBAYASHI Toshihiko / IWASA Kenji

Huge amounts of diverse data can be collected in semiconductor manufacturing processes accompanying the ongoing miniaturization of semiconductor devices, the increase in the number of process steps, and the advances in equipment. Demand has therefore been growing for greater efficiency and sophistication in quality analysis work using artificial intelligence (AI) technologies. In order to maximize the effectiveness of AI, a platform capable of centrally managing data while maintaining relationships among manufacturing process steps, equipment, and conditions is essential. However, flexible integration of these data has been difficult to achieve using conventional relational databases (RDBs).

To address this issue, the Toshiba Group has developed and launched Meister SemiSmartDF, a unified manufacturing data management and utilization platform integrating its accumulated semiconductor manufacturing know-how and proprietary AI technologies. We have applied the platform to actual manufacturing environments and confirmed that it can reduce the time required for quality analysis work from 4.2 hours to 0.5 hours per person per day, while also contributing to the extraction of primary defect-causing factors from among a large number of factors through sparse modeling.

FUNAE Kouki / GOYA Taku / ENDO Kiyokazu

Equipment-intensive industries have recently been facing critical issues, including aging equipment and labor shortages. As a result, there is growing demand for advanced decision-support solutions to implement equipment maintenance and renewal with limited resources. However, as operational technology (OT) data on equipment condition and information technology (IT) data on maintenance histories, plans, and costs are managed by individual departments, totally optimized decision-making based on OT/IT data has not been achieved to date.

Toshiba Corporation has responded to this situation by devising a solution integrating OT/IT data in a digital twin, making it possible to support decision-making in equipment maintenance and operation. This solution will assist in achieving total optimization of equipment maintenance and capital investment plans based on risk-based maintenance (RBM) with higher accuracy than conventional time-based maintenance (TBM) and condition-based maintenance (CBM).

SHIROTA Yusuke / WAKAMATSU Tomohiro / KANAI Tatsunori / NAKATA Kouta

Digital receipts have recently attracted attention as big data effective for understanding consumers’ purchasing behavior. To extract market structures and latent needs from large volumes of diverse data, advanced artificial intelligence (AI) applications are increasingly being introduced as alternatives to conventional evaluation methods. However, challenges remain, including difficulties in understanding market structures when products are regrouped according to purchasing behavior rather than conventional category systems, as well as in rapidly and objectively interpreting analysis results.

To address these issues, Toshiba Corporation has developed receipt informatics technology using composite AI specialized for purchase big data. This technology makes it possible to generate previously unobtainable high-value market insights by automatically extracting the need axes through preference-clustering AI. By integrating large language model (LLM)-based feature summarization to automate the appropriate cluster labeling, the technology also facilitates comprehensive understanding of overall market structures.

YAMAGUCHI Taihei

Preparation for infectious disease outbreaks, rising healthcare costs associated with an aging society, and the increase in chronic diseases have become issues of vital importance. To address these social issues, a platform supporting preventive healthcare and improving healthcare measures through the continuous integration and analysis of real-world data and the application of the resulting knowledge to medical and healthcare fields is essential.

Toshiba Corporation has constructed a platform based on a corporate cohort of approximately 19 000 domestic employees of the Toshiba Group by continuously and individually integrating health and medical data with workplace behavioral data, such as attendance records and purchasing data, and conducted joint research with multiple research institutions. A study on adverse reactions following COVID-19 vaccine booster shots enabled exploratory analysis of genetic factors by combining data accumulated on the platform with additional questionnaire data. Another study involving a randomized controlled trial (RCT) of sodium reduction clarified both the feasibility of and the challenges associated with implementing personalized prevention. These studies demonstrate the potential of the platform based on the corporate cohort to support both preventive healthcare services in normal times and rapid factor analysis during times of crisis.

SAITO Minoru / ENDO Kotaro

With industries facing various problems associated with the ease of digital data forgery accompanying the recent progress of generative artificial intelligence (AI), ensuring data authenticity in business-to-business (B2B) transactions has become an important issue. Blockchain, a distributed ledger resilient to tampering, is a key technology for guaranteeing data integrity without the need for third parties.

The Toshiba Group has developed DNCWARE Blockchain+ (hereafter referred to as “BC+”), a blockchain for enterprises that ensures data authenticity through multi-layered mechanisms including tamper detection, business rule enforcement, and confidentiality protection, and has launched BC+ for logistics management and municipal electronic contract systems. We are currently working on the verification of an automated payment system using stablecoins compliant with the inter-blockchain communication protocol (IBC) and an automated transaction system in collaboration with AI agents.

MIHARA Isao / KOMATSU Misaki

In response to the expanding use of artificial intelligence (AI) systems, ensuring safety and reliability of their operation through appropriate risk management has become an issue of critical importance. To realize trustworthy AI systems, Toshiba Corporation has taken the initiative in establishing AI governance and engaging in AI risk management. In particular, continuous evaluation and countermeasures are required to address AI security risks. However, conventional assessment methods are insufficient in practical environments that use external application programming interfaces (APIs) and existing models, where access to internal information is restricted.

To rectify this issue, we have developed a black-box AI risk assessment method that assesses risks through input–output behavior analysis by treating AI systems as black boxes. This method enables continuous risk management of AI systems, including those in operation.

Feature Articles

HIRANO Itsuki / SHIOKAWA Miyuki / GOTANDA Takeshi

Renewable energy generation, such as photovoltaic (PV) power generation, is a focus of high expectations as a future major power source. In particular, demand has arisen for solar cells with higher efficiency and longer lifetimes than ever before in line with the widespread adoption of PV systems.

In response to this market demand, Toshiba Corporation is vigorously promoting the practical realization of perovskite/silicon tandem solar cells. In 2025, we developed a tandem solar cell achieving a power conversion efficiency (PCE) of 31.3%, higher than that of dominant monocrystalline silicon solar cells, by optimizing the device design, including current matching between perovskite and silicon layers, as well as the passivation process. We have conducted long-term light soaking tests and outdoor tests, verifying that the cell exhibits highly stable performance by suppressing degradation through optimization of transparent electrode sputtering conditions. We are continuously engaged in development through evaluations under a wide variety of solar irradiation conditions, such as those associated with building facades and elevated structures, as well as analyses and field tests related to the risk of lead leakage in the event of glass breakage.

MORI Yoshinori / YOSHIDA Takeshi / SAITO Takuya

Vacuum interrupters are commonly used in switchgear supporting electric power equipment. In particular, those with silver-tungsten carbide (Ag-WC) electrical contacts feature high breaking and low surge performance. However, as vacuum contactors are among the main types of switchgear required for more frequent switching operations than vacuum circuit breakers, vacuum interrupters must not only achieve high mechanical durability but also suppress long‑term changes in contact resistance.

Against this background, Toshiba Corporation has focused on adhesion phenomena that tend to occur during the opening and closing of electrical contacts in a vacuum, and has clarified that changes in the surface properties of electrical contacts affect the mechanical durability and stable contact resistance of vacuum interrupters. Evaluation experiments on the long-term performance of actual equipment using newly developed Ag-WC electrical contacts based on this study have confirmed that they achieve stable contact resistance for a prolonged period.

ITADERA Tatsuyoshi / MATSUURA Masakazu / TARUKI Hisayuki

Linear image sensors have been widely used not only in office automation systems, including multifunctional peripherals (MFPs), but also in image inspection systems for detecting defects and faulty items in industrial and agricultural applications. To enable rapid acquisition of high-resolution images, linear image sensors for visual inspection must be able to address external analog front-end (AFE) components, such as timing generators, drive circuits, and output amplifier circuits, in addition to enhancing sensor performance.

Toshiba Electronic Devices & Storage Corporation has developed the TCD2400DG charge-coupled device (CCD) linear image sensor, which achieves a line rate of 22.7 kHz, approximately double that of conventional products, by enhancing the sensitivity of CCD pixels, accelerating charge transfer, improving the signal-to-noise (S/N) ratio, and incorporating part of AFE circuits that were externally mounted in conventional products. The device was released in December 2025.

Frontiers of Research & Development

AI Agent Using Generative AI to Improve Efficiency of Setup and Operation of Automatic Image Inspection Systems

*Company, product, and service names appearing in each paper include those that are trademarks or registered trademarks of their respective companies.

Vol. 81, No. 4, July 2026

Index