Technical Talks
December 2021
Design Techniques to Mitigate Radiation Effects on Image Sensors
- Speaker: Rico Jossel M. Maestro
- Date: 15 December 2021
Abstract
Image sensors are being used more than ever in different applications: from capturing our daily lives with our smartphones, to capturing how other planets look like. Some of these applications expose the sensor in harsh environments such as the presence of radiation. As such, aside from its typical specifications, the image sensor system should also be designed with reliability and longevity in mind. This talk will provide an overview on image sensor systems and on how radiation affects its performance. Then, design techniques will be discussed to ensure that the system will still be reliable even in the presence of radiation.
An Energy-Efficient Unsupervised Online Learning Classifier for Seizure Detection
- Speaker: Adelson Chua
- Date: 9 December 2021
Abstract
Epilepsy is a serious neurological disorder affecting around 50 million people worldwide and is usually characterized by recurrent seizures. Implantable devices that record neural activity and detect seizures have been adopted to issue warnings or trigger neurostimulation to suppress epileptic seizures. Typical seizure detection systems rely on high-accuracy offline-trained machine learning classifiers that require manual retraining when seizure patterns change over long periods of time. For an implantable seizure detection system, a low-power, at-the-edge, online learning algorithm can be employed to dynamically adapt to the neural signal drifts, thereby maintaining high accuracy without external intervention. In this talk, I will be presenting an unsupervised online learning classifier using stochastic gradient descent and logistic regression. After an initial offline training phase, continuous online unsupervised classifier updates are applied in situ, which improves sensitivity in patients with drifting seizure features. The classifier was tested on two human electroencephalography (EEG) datasets, where it was able to achieve an average sensitivity of 97.5% and 97.9% respectively, and maintaining <1.2 false alarms per day. The classifier was fabricated in TSMC’s 28 nm process occupying 0.1 mm2 and achieves 1.5 nJ/classification energy efficiency, which is at least 24x more efficient than state-of-the-art.