人工智能 AI

DIGITIMES "AI on Air: Computex Tech Forum." As we approach Computex 2026, the narrative within the global supply chain is definitively shifting. AI is no longer just a software marvel confined to cloud servers; it is actively reconstructing hardware architectures, pushing our energy grids to their physical limits, and taking shape in the real world through Autonomous Vehicles (AVs) and Humanoid Robotics.

Drawing from the insights of industry leaders and star analysts at the forum, here is a deep dive into the three macro-trends that will define the next phase of the AI ecosystem.

1. The Great Pivot to Inference & The Rise of Enterprise AI Agents

We are transitioning from the "Arms Race of Training" to the "Era of Inference." A prime indicator is Anthropic's Claude, which hit a $50 billion valuation in May 2026. Its Annual Recurring Revenue (ARR) in the enterprise sector is showing phenomenal traction, holding over 60% of the enterprise market share while operating with much higher token efficiency compared to competitors.

As Agentic AI becomes the standard, the hardware bottleneck is no longer just the AI accelerator (GPU/ASIC) itself. The friction has moved to the CPU, memory bandwidth, and system networking.

• The New Metric: Hardware success is now defined by Tokens/Second optimization.

• Architectural Innovation: Innovators like Groq (with their LPU) are specifically tackling the "first-token" compute bottleneck and decoding memory hurdles. By utilizing highly optimized architectures to process Mixture of Experts (MoE) parameters, they are paving the way for instantaneous AI responses. Meanwhile, advanced packaging constraints (like Samsung's CoWoS/HBM yield rates) remain a critical variable in global output.

2. Hitting the Power Wall & The Pragmatic Reality of Liquid Cooling

Compute expansion is fundamentally constrained by electricity. As power becomes the ultimate bottleneck for data centers, thermal management has moved from a facility issue to a strategic boardroom topic.

Insights shared by YuanTai Tech highlighted the pragmatic realities of liquid cooling deployments:

• Direct-to-Chip (D2C) is the Current King: Utilizing 35–45°C inlet water paired with cooling towers effectively eliminates the need for power-hungry traditional chillers. Because D2C is performing so well and driving PUE (Power Usage Effectiveness) down significantly, the immediate market urgency for Immersion Cooling has somewhat stabilized.

• Engineering the Details: The real-world success of liquid cooling hinges on two critical factors: Negative-pressure cabinets (ensuring water absolutely never leaks onto servers) and strict micro-channel filtration. Without proper filtration before the Cooling Distribution Unit (CDU), the tiny micro-channels on cold plates will quickly clog with debris, leading to catastrophic system failures.

3. Physical AI: The "World Model" Revolution

AI is gaining a body. In 2026, AVs are projected to account for 65% of shipments, with ADAS/AV chip penetration growing exponentially toward 2030. But the real breakthrough is architectural.

Autonomous driving is abandoning the traditional modular design (Perception ➡️ Localization ➡️ Planning ➡️ Control) in favor of End-to-End Deep Learning combined with "World Models." To navigate the physical world, AI must possess:

• Spatial Cognition: Understanding 3D environmental structures, the size of pedestrians, obstacles, and viable lane changes in real-time.

• Temporal Cognition: Predicting how the physical world will change over time. It’s not just about seeing a pedestrian; it’s about calculating their trajectory over the next 5 seconds to preemptively brake or steer.

Coupled with the explosive growth of humanoid robots (with companies like Tesla, Boston Dynamics, Figure AI, and emerging Chinese robotics firms accelerating development), Physical AI is the ultimate frontier.

Strategic Takeaway: The next decade belongs to organizations that can bridge the gap between algorithmic intelligence and physical execution, all while managing the strict constraints of power and heat. The Computex floor this year will be the ultimate proving ground for these solutions.

Would like to know more...reach the full report from Digitimes (Reminder: Charged Report---and I am not seller XD )https://www.digitimes.com.tw/seminar/TechForum_20260520/

1. Edge Intelligence: Qualcomm and Realtek demonstrated complex reasoning at the terminal level, enabling robots with real-time adaptability without relying on unstable cloud connections.
2. Storage Architecture Rescaling: Phison aiDAPTIV+ technology ingeniously addresses the severe memory shortage of 2026, allowing SMEs to train private AI models on constrained budgets.
3. Safety and Trust: As emphasized by WHALE Forensics, safety is no longer optional but a requirement for entering "Non-Red Supply Chains." Taiwan's AI Basic Act, harmonized with international standards, provides the legal foundation for exporting "Trust."
4. Sustainable Manufacturing: From energy-efficient architectures in R&D to the "Showroom-to-Factory" carbon reduction model at TMTS 2026, Taiwan is demonstrating how to transform ESG into market competitiveness.

Ref: Chiou, Y.-L., & Chen, H.-L. (2025, November 21–22). From automation to intelligence: Analyzing patent trends to reveal the evolution and healthcare applications of TCM pulse diagnosis technology. 16th International Conference on Educational Innovation (ICEI), National Tsing Hua University, Hsinchu, Taiwan.

Abstract

This study analyzes 850 international patents (1986-2025) to map the technological modernization of TCM pulse diagnosis, traditionally limited by subjectivity. We identify a three-stage evolution: Hardware (objective signal acquisition), Analytical (feature extraction), and Intelligent (AI-driven diagnostics). Key findings identify a strategic inflection point towards deep learning models and cloud-based platforms, enabling transformative applications in integrative medicine, personalized remote healthcare, and standardized medical education. However, realizing this potential is contingent on overcoming critical implementation hurdles, including data standardization, regulatory validation, and knowledge integration.

Keywords: Pulse Diagnosis, Patent Analysis, Traditional Chinese Medicine (TCM), Remote Healthcare, AI Collaboration Interdisciplinary Education

1. Introduction

TCM pulse diagnosis, a cornerstone method, traditionally relies on practitioner subjectivity, hindering standardization and integration into modern healthcare (Tang et al., 2024). Responding to these limits, the field shifted technologically from "automation" (objective hardware sensors capturing pulse data) to "intelligence" (AI-driven analytics enabling automated interpretation and prediction). This study uses international patent analysis to map this evolution, guided by three research questions:

1. What are the patent development trends associated with integrating automation and intelligence in TCM pulse diagnosis technology?

2. How have core technologies evolved within the field of automated and intelligent TCM pulse diagnosis?

3. From a patent trend perspective, what are the potential healthcare applications of automated and intelligent TCM pulse diagnosis?

This paper first details the patent-centric methodology used to analyze the technology landscape. It then presents a three-stage framework of technological evolution, analyzes the key findings and applications derived from the patent data, and concludes with a discussion of implementation challenges and strategic recommendations for stakeholders in this innovative field.

2. Research Methodology

This study used a patent-centric framework. Data was sourced from the GPSS (TIPO) , covering patents filed between 1980-2025, resulting in a final dataset of 850 relevant patents (1986-2025). A multi-layered Boolean search strategy was used, as Cui, J., & Xu, J. T. (2025) mentioned, targeting core concepts (e.g., pulse diagnos*, Cun Guan Chi), technologies (e.g., sensor*, AI), and TCM context. The analysis involved a dual method: quantitative bibliometrics for trends and qualitative review of patent specifications to detail hardware (e.g., sensor technology) /software (e.g., signal processing and AI algorithms) trajectories.

3. The Three-Stage Evolution of Pulse Diagnosis Technology

The patent analysis reveals a clear, three-stage evolutionary framework for TCM pulse diagnosis technology. This progression maps a logical journey from solving the fundamental challenge of objective hardware measurement to achieving the sophisticated goal of automated, intelligent interpretation.

3.1. Stage 1: Hardware Evolution - The Quest for Objective Signal Acquisition

This stage focused on replacing subjective palpation with quantifiable data using various sensing technologies.

3.1.1. Pressure Sensors: Mimicking palpation, evolved to MEMS arrays for multi-point measurement (Cun, Guan, Chi), capturing pulse amplitude/depth and width (e.g., CN200610073864.1, CN201310126155.5, US17121877).

3.1.2. Optical Sensors: Non-contact method evolved to wearable PPG sensors measuring blood volume changes, trending toward consumer devices (e.g., CN201810234938.8, US09341248).

3.1.3. Piezoelectric and Air-Based Systems: Piezoelectric sensors in inflatable cuffs use controlled air pressure to simulate consistent force, capturing pulse vibrations (e.g., US14408543, CN202310015477.6).

3.1.4. Multi-Modal Sensing: Integrates multiple sensor types (e.g., pulse, blood oxygen, ECG) for a holistic physiological profile and richer data (e.g., CN105030195A).

3.2. Stage 2: Analytical Evolution - From Raw Signals to Meaningful Features

With reliable hardware established, focus shifted to signal processing to refine noisy raw data (from movement, breathing, etc.) and extract quantifiable features.

3.2.1. Signal Preprocessing: Improves signal quality using methods like wavelet transforms and filtering to remove noise and baseline drift (e.g., CN201810795126.0).

3.2.2. Time-Domain Feature Extraction: Directly analyzes waveform characteristics, like PPG systolic (PWSP) and diastolic (PWDP) peaks, for shape/rhythm metrics (e.g., CN202310878328.2).

3.2.3. Frequency-Domain Feature Extraction: Converts time-series data to the frequency domain using FFT or wavelet analysis to analyze power distribution, indicating physiological states (e.g., CN202110127009.9).

3.2.4. Bridging to Western Cardiology: Connects extracted features (e.g., Pulse Transit Time/PTT, Augmentation Index/AIx for arterial stiffness) to Western biomarkers, aiding validation and integration (e.g., US15901375).

3.3. Stage 3: Intelligent Evolution - The Rise of AI in Diagnostics

This current stage shifts from descriptive analytics to predictive intelligence using feature-rich datasets. It involves widespread use of Machine Learning (ML) and Deep Learning (DL) for automated interpretation and diagnosis.

3.3.1. Pulse Type Classification: Neural networks automatically identify specific TCM pulse types (e.g., "wiry pulse"). Advanced hybrid models like CNN-LSTM predict characteristics such as pulse strength (e.g., CN202410083997.5, CN201810247234.4).

3.3.2. Advanced Prediction Models: AI performs complex predictions beyond classification, such as pulse phase prediction (e.g., CN202110835433.9).

3.3.3. Intelligent Control Systems: AI enhances hardware, like in bionic pulse diagnosis equipment where it simulates doctor's actions and precisely controls pressure for improved signal acquisition (e.g., CN202311759950.8).

3.3.4. System Architecture Trends: Drives a shift toward cloud-based platforms, enabling new business models and data analytics opportunities (e.g., CN201810054555.2).

The maturation of these intelligent systems is the critical enabler for the new generation of healthcare solutions and applications discussed in the following section.

4. Research Findings and Applications

This section directly addresses the study's research questions by presenting the high-level patent trends, analyzing the Technology-Efficacy Matrix to map core innovations to diagnostic functions, and outlining the potential healthcare and educational applications that emerge from this technological evolution.

4.1. Patent Development Trends (Answering RQ1)

Analysis of 850 patents shows a clear shift towards AI integration ML/DL models such as Support Vector Machines (SVM), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) networks. A parallel trend is the adoption of cloud platforms, enabling remote diagnostics and scalable business models.

4.2. Technology-Efficacy Matrix Analysis (Answering RQ2)

A Technology-Efficacy Matrix, showed as Fig.1, mapping core technologies to functions using patent counts, reveals heavy R&D concentration on applications.

4.3. Potential Healthcare and Educational Applications (Answering RQ3)

The evolution from automation to intelligence enables impactful applications: (1) Integration of TCM and Western Medicine: multi-modal data bridges medical paradigms by correlating TCM waveforms with Western biomarkers (e.g., blood oxygen, ECG), facilitating holistic, data-driven patient assessment. (2) Personalized and Preventive Healthcare: Wearables enable continuous self-monitoring, early warnings, and a shift to proactive health management. (e.g., TW108117679). (3) Telemedicine and Remote Health Management: Intelligent remote diagnostics expand access, especially for aging/remote populations. Systems (e.g., TW110114238) allow home data capture and cloud analysis, fostering patient-centric care. (4) Interdisciplinary Education and Standardization: AI serves as a collaborative partner in medical education. Digital pulse databases/simulators provide standardized training resources, objectifying learning and promoting interdisciplinary understanding. While the potential value of these applications is immense, successful implementation requires overcoming several real-world challenges, which are addressed next.

5. Discussion

Realizing the potential of intelligent pulse diagnosis requires overcoming key implementation challenges related to data, regulation, and knowledge integration for widespread adoption.

5.1. Key Implementation Hurdles

• Data Standardization: Lack of large, standardized public datasets hinders AI training, validation, and comparison.

• Regulatory and Clinical Hurdles: Rigorous validation for safety/accuracy and navigating approval pathways (e.g., FDA) are significant barriers.

• Knowledge Integration: Translating qualitative TCM practitioner knowledge into quantitative algorithms remains a core challenge.

5.2. Strategic Recommendations

To overcome these hurdles and accelerate the adoption:

1. Establish Unified Standards: Create public benchmark datasets, essential for Stage 3 AI generalizability.

2. Promote User-Centric and Human-Centered Design: To maximize the value proposition for applications in remote medicine and education, development must prioritize devices that are non-invasive, comfortable, and intuitive. A positive user experience for both practitioners and patients is critical for ensuring high-quality data collection and compliance, thereby strengthening the entire data-to-diagnosis pipeline.

3. Develop Robust Ethical Frameworks: Address data privacy, cybersecurity, and responsible AI use, especially for Stage 3 cloud platforms.

4. Adopt Differentiated Patent Positioning: Focus IP on proprietary algorithms and system solutions beyond just hardware for Stage 3 competitiveness. Lies in proprietary algorithms, integrated system-level solutions, and unique cloud-based service architectures that create significant barriers to entry.

By strategically addressing these challenges, the field can build a solid foundation for the clinical integration and commercial success of this transformative technology.

6. Conclusion

This analysis of the patent and academic landscape reveals a clear and compelling technological narrative: the evolution of Traditional Chinese Medicine pulse diagnosis from a subjective art to an objective, data-driven science. The study has traced its definitive three-stage progression from hardware automation aimed at capturing a reliable signal, through an analytical phase focused on extracting meaningful features, to the current era of AI-driven intelligence, where systems can interpret and classify complex pulse patterns with increasing sophistication. This technological maturation is unlocking transformative potential across the healthcare spectrum by creating a data-driven bridge for integrating TCM and Western medicine, empowering individuals with tools for personalized care, and enabling new models of telemedicine. Realizing this future is not inevitable; it is contingent upon a concerted, cross-disciplinary effort to resolve the fundamental challenges of data standardization, clinical validation, and ethical governance. Only then can this ancient diagnostic art be fully realized as a data-driven science for 21st-century healthcare.

7. References

Beijing Institute Of Technology. (2023). Deep learning pulse signal automatic acquisition system and method (Patent No. CN202311759950.8). China National Intellectual Property Administration.

Beijing Songtou Technology Co., Ltd. (2018). Intelligent traditional Chinese medicine pulse diagnosis method and system (Patent No. CN201810247234.4). China National Intellectual Property Administration.

Beijing University Of Chinese Medicine. (2013). Automatic three-portion and nine-pulse-taking information acquisition and recognition device and method of MEMS hydraulic transmission touch (Patent No. CN201310126155.5). China National Intellectual Property Administration.

Cui, J., & Xu, J. T. (2025). Application status and prospects of artificial intelligence information technology in the modernization research of TCM four diagnostic methods. Journal of Traditional Chinese Medicine. Advance online publication. https://doi.org/10.16305/j.1007-1334.2025.z20240903008

Dr. Advice Technology Corporation Limited. (2019). TCM pulse condition detection bracelet (Taiwan Patent No. TW108117679). Taiwan Intellectual Property Office.

Faceheart Inc. (2021). System and method adopting contactless physiological signal measuring technique to accomplish remote TCM pulse diagnosis (Taiwan Patent No. TW110114238). Taiwan Intellectual Property Office.

Hangzhou Zhangyu Technology Co., Ltd. (2023). Air bag type bionic finger pulse feeling instrument (Patent No. CN202310015477.6). China National Intellectual Property Administration.

Hong Kong Polytechnic University. (2006). Three portions and nine pulse-takings pulse condition detector of pulse condition sensor of herbalist doctor, and pulse condition detection method (Patent No. CN200610073864.1). China National Intellectual Property Administration.

Institute Of Microelectronics Of Chinese Academy Of Sciences. (2021). Pulse wave acquisition pressure control method, device and system (Patent No. CN202111553703.3). China National Intellectual Property Administration.

International Business Machines Corporation. (2018). Heart rate and blood pressure monitoring biosensors (U.S. Patent Application No. 15/901,375). U.S. Patent and Trademark Office.

Ji Hua Laboratory. (2024). Pulse power grade prediction method based on CLLSR hybrid model (Patent No. CN202410083997.5). China National Intellectual Property Administration.

Korea Institute Of Oriental Medicine. (2013). Pulse measuring device and method of pulse measurement therewith (U.S. Patent No. 14/408,543). U.S. Patent and Trademark Office.

Qianhai Anycheck Information Technology Company; Shenzhen E-Techco Information Technology Co., Ltd. (2018). Remote processing system and method for face images and pulse manifestation of traditional Chinese medicine (Patent No. CN201810054555.2). China National Intellectual Property Administration.

Seiko Epson Corporation. (1999). Method of detecting pulse wave, method of detecting artery position, and pulse wave detecting apparatus (U.S. Patent No. US09341248). U.S. Patent and Trademark Office.

Shanghai Taiyi Health Technology Co., Ltd. (2018). Physique detection method based on traditional Chinese medicine pulse condition analysis (Patent No. CN201810795126.0). China National Intellectual Property Administration.

Shanghai University Of Traditional Chinese Medicine. (2021). Pulse wave physiological information extraction method (Patent No. CN202110835433.9). China National Intellectual Property Administration.

Shenyang Tianren Heyi Technology Co., Ltd. (2023). Pulse taking method for pulse diagnosis instrument (Patent No. CN202310878328.2). China National Intellectual Property Administration.

South China Normal University. (2018). Pulse condition analysis method based on PPG signals and image enhancing (Patent No. CN201810234938.8). China National Intellectual Property Administration.

Taiwan Intellectual Property Office. (2025). Global Patent Search System. Retrieved July 4, 2025, from https://gpss.tipo.gov.tw/

Tang, Q., Chen, Y.-k., Liu, S., Wang, J., Zhang, L., Qu, H., & An, H. (2024). Pulse signal may be a key bridge connecting cardiology and pulse diagnosis of TCM. Heliyon, 10(16), e36785. https://doi.org/10.1016/j.heliyon.2024.e36785

Tsinghua Shenzhen International Graduate School. (2021). Pulse data classification model establishment device, classification identification device and measurement system (Patent No. CN202110127009.9). China National Intellectual Property Administration.

Wellness Allied Inc. (2020). Pulse device and method to instruct treatment with pulse device (U.S. Patent Application No. 17/121,877). U.S. Patent and Trademark Office.

Zhao, Q., Wang, L., Wang, L., Liu, C., & Ding, X. (2015). Pulse-taking health care data acquisition system and pulse-taking health care system (Patent No. CN105030195A). China National Intellectual Property Administration.