A5046163518- Advanced Memory and Neural Computing
- Ferroelectric and Negative Capacitance Devices
- Neuroscience and Neural Engineering
- Ferroelectric and Piezoelectric Materials
- ECG Monitoring and Analysis
- Machine Learning and ELM
- Brain Tumor Detection and Classification
- Acoustic Wave Resonator Technologies
- CCD and CMOS Imaging Sensors
- Speech and dialogue systems
- Neural Networks and Applications
- Transition Metal Oxide Nanomaterials
- EEG and Brain-Computer Interfaces
- Blind Source Separation Techniques
- Photoacoustic and Ultrasonic Imaging
Nanyang Technological University
2020-2025
Institute of Microelectronics
2021-2023
Agency for Science, Technology and Research
2021-2023
The miniaturization and real time imaging capability have always been the desired properties of photoacoustic (PAI) system, which unlocked vast potential for personalized healthcare diagnostics. While quality resolution in such systems are inferior due to physics system volume constraints, limited its wide deployment application. This paper proposes a novel platform enhance handheld PAI time, integrating MultiResU-Net enhancement algorithm with Ferroelectric random-access memory (FeRAM)...
Real-time speech interaction, serving as a fundamental interface for human-machine collaboration, holds immense potential. However, current open-source models face limitations such high costs in voice data collection, weakness dynamic control, and limited intelligence. To address these challenges, this paper introduces Step-Audio, the first production-ready solution. Key contributions include: 1) 130B-parameter unified speech-text multi-modal model that achieves understanding generation,...
In this work, by experiments and material calculations, the steep polarization switching leakage in ferroelectric Al <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.7</inf> Sc xmlns:xlink="http://www.w3.org/1999/xlink">0.3</inf> N are investigated. The calculations suggest that tight distribution of coercive field is attributed to highly uniform with well-aligned domains. electron emission hopping assisted vacancies layer dominate current. For...
This paper presents a parasitic-aware modeling method for fast and accurate simulation of Processing-in-Memory (PIM) neural network (NN) implemented in resistive memristor crossbar array. work proposed an efficient line resistance estimation model named <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\gamma $ </tex-math></inline-formula> -compact model, the associated NN training scheme that takes impact...
In this work, a non-idealities aware software-hardware co-design framework for deep neural network (DNN) implemented on memristive crossbar is presented. The device level non-ideal factors such as conductance variation, nonuniform quantization levels, device-to-device variation and programming failure probability are included in the model. At array level, impact of line resistance sneak path considered using new fast accurate estimation non-linearity offset peripheral circuits also...
This work presents an edge-AI system built on capacitive ferroelectric random-access memory (FeRAM) crossbar array, which is compatible with CMOS backend-of-line (BEOL) fabrication process. A novel circuit and a ternary mapping technique are proposed. Compared to the conventional binary representation, proposed improves storage efficiency exponentially in weight resolution. The feasibility of neuromorphic computing implemented FeRAM array explored speech command classification task....
In this paper, the performance analysis of convolutional neural network (CNN) with multi-level memristor crossbar is presented. Multi-level used to implement Vector-Matrix Multiplication (VMM), which most computationally intensive step in CNN algorithm. A procedure convert a classical model floating-point accuracy weights finite-bit implemented multilevel The impacts levels, line resistance and array size VMM calculation classification are analyzed details. As an example, one converted...
PoolFormer is a type of neural network architecture that abstracted from Transformer where the computationally heavy token mixer module replaced with simple pooling function. This paper presents memristor-based modeling and training framework for edge-AI applications. To fit implementation on resistive crossbar array, original structure further optimized by replacing normalization operation hardware friendly scaling operation. In addition, non-idealities RRAM device to array level as well...
This paper presents a parasitic-aware modelling approach called αβ-matrix model for the simulation of neural network (NN) implemented with memristor crossbar array. The line resistance, which is key parasitic in array analyzed and incorporated into model. proposed method estimates resistance IR drop computation complexity O(mn), contrast to O(m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> n ) required by classical matrix based...
This paper presents an electrocardiogram (ECG) signal classification method using binary CNN implemented on RRAM crossbar arrays. A new structure is proposed to provide input-dependent references for adaptive readout quantization. Hence, weights can be represented with just one cell, instead of the conventional differential leading reduction size by half. Furthermore, impacts nonidealities mitigated nonidealities-aware training and in-situ compensation. On other hand, bandpass filter (BPF)...