Interviews with Faculty Researchers

– Interview with Prof. Han Suting
Associate Professor, Department of Applied Biology and Chemical Technology

The emergence of brain-inspired (neuromorphic) computing offers a promising route to overcome the limitations of conventional von Neumann architectures in artificial intelligence (AI). While traditional systems separate memory and computation—resulting in high energy consumption—the human brain integrates these functions efficiently within a compact structure. Addressing this gap, Prof. Han Suting’s research focuses on memristor-based neuromorphic hardware, enabling AI systems that more closely emulate biological intelligence.

Memristors are two-terminal devices capable of both storing and processing information, making them ideal for in-memory computing. By continuously adjusting their conductance in response to electrical signals, they mimic the adaptive behaviour of biological synapses. This allows computation to occur directly within memory, eliminating costly data transfer and significantly improving speed and energy efficiency.

Through crossbar array architectures, memristor systems perform vector–matrix multiplication—a core neural network operation—in a highly parallel manner. This contrasts with the sequential processing of conventional systems, enabling faster and lower-power computation while supporting synapse-like functionality in hardware.

Prof. Han’s work also incorporates biologically inspired learning mechanisms, particularly spike-timing-dependent plasticity (STDP), enabling adaptive weight updates in memristor arrays. This supports the development of spiking neural networks (SNNs), which more closely resemble natural neural systems.

At the materials level, her research explores hybrid perovskite and organic materials, where ion migration enables precise conductance modulation. By optimizing crystallinity and introducing passivation layers, her team improves device performance, stability and scalability.

Beyond theory, these technologies show strong potential in real-world applications. Flexible, wearable memristor-based systems have been developed for in-sensor computing, integrating sensing, memory, and processing into a single platform. Such systems enable intelligent responses to environmental stimuli, supporting low-power, real-time AI in areas such as healthcare and robotics.

Looking ahead, her work extends to human–machine interfaces, including assistive technologies for visual impairments, reflecting a broader vision of compact, brain-like, energy-efficient systems. Together, these efforts position Prof. Han’s research at the forefront of memristor-based neuromorphic hardware, bridging the gap between silicon systems and biological intelligence.

受人腦所啟發：以憶阻器類神經形態硬件帶動高效能人工智能

– 韓素婷教授專訪
應用生物及化學科技學系副教授

類神經形態運算的興起，為人工智能中的傳統馮·諾依曼架構帶來了突破。傳統計算系統將「記憶」與「運算」分離，產生了大量數據搬移與能耗問題；相比之下，人類大腦能高效地同時運行兩者，展現出極高能效與適應能力。對此，韓素婷教授專注研究有關憶阻器的類神經形態硬件，以令人工智能系統能更貼近人類的運作模式。

憶阻器是一種具備記憶功能的雙端元件，能同時進行資訊儲存與處理，非常適合應用於存算一體架構。它可根據電訊號不斷調節電導，模擬生物之間的溝通方式。將運算直接嵌入記憶體中，系統可大幅減少數據在記憶體與處理單元之間的傳輸功耗，大幅提升運算速度。

透過交叉陣列架構，憶阻器系統能以高度並行方式執行神經網絡的核心運算——向量–矩陣乘法。相比傳統序列式運算方式，其速度更快、功耗更低，能在硬件層面模擬生物溝通。

韓教授的研究亦引入了啟發自生物的學習機制，包括脈衝時序依賴可塑性技術，使憶阻器陣列能根據輸入脈衝的時間關係自動調整權重，進一步讓其脈衝神經網絡行為更接近自然神經系統的運作模式。

團隊亦重點探索混合鈣鈦礦與有機材料，利用離子遷移機制精確調控電導率。透過改進晶體結構及引入界面鈍化層，進一步提升憶阻器的性能、穩定性及可規模性。

這些技術在實際應用中潛力巨大。團隊已開發出柔性可穿戴憶阻器系統，用於傳感器內運算，以將感測、記憶與運算整合於一體，並對四周環境作出反應，適用於醫療與機械人等低功耗領域。

展望未來，韓教授的研究將延伸至人機介面，包括針對視覺障礙的輔助技術，以發展出體積更小、性能與人腦更相近的智能系統，令其成為未來矽基系統與生物智能的核心科技。