Aakash Patel
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Aakash Patel has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Dwaipayan Biswas
IMEC, Leuven, Belgium
2 shared publications - Ivan Dario Castro
IMEC, Leuven, Belgium
2 shared publications Julien Ryckaert
Interuniversitair Micro-Elektronica Centrum (IMEC), Leuven, Belgium
2 shared publications- Khakim Akhunov
Department of Information Engineering and Computer Science, University of Trento
2 shared publications - Tom Torfs
IMEC, Leuven, Belgium
2 shared publications Bernhard Laufer
Institute of Technical Medicine , Furtwangen University , Villingen-Schwenningen , Germany
1 shared publication
Latest Publications
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 25 (1) , 230-234
Wafer-Scale GPU Memory Pool With In-Package Optics for Enhanced Capacity and Bandwidth
The escalating demands of modern AI applications push the limits of conventional GPU memory systems, where high-bandwidth memory (HBM) architectures face critical challenges in scalability and bandwidth efficiency. We propose a wafer-scale system integrating photonic interconnects to decouple GPU compute from memory, addressing capacity and bandwidth limitations of conventional HBM-based architectures. We develop an analytical framework, HBM-pool, blending in-package optical I/O (OIO) parameters with GPU memory traffic for high-density nodes. The system-technology co-optimization (STCO) study shows that 300 mm wafer-scale disaggregation with OIO can achieve a 4× increase in memory capacity and reduce bandwidth bottlenecks up to 8× compared to a 2.5D electrically packaged node, enabling larger training/inference and improving latency.
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 25 (1) , 230-234
Wafer-Scale GPU Memory Pool With In-Package Optics for Enhanced Capacity and Bandwidth
The escalating demands of modern AI applications push the limits of conventional GPU memory systems, where high-bandwidth memory (HBM) architectures face critical challenges in scalability and bandwidth efficiency. We propose a wafer-scale system integrating photonic interconnects to decouple GPU compute from memory, addressing capacity and bandwidth limitations of conventional HBM-based architectures. We develop an analytical framework, HBM-pool, blending in-package optical I/O (OIO) parameters with GPU memory traffic for high-density nodes. The system-technology co-optimization (STCO) study shows that 300 mm wafer-scale disaggregation with OIO can achieve a 4× increase in memory capacity and reduce bandwidth bottlenecks up to 8× compared to a 2.5D electrically packaged node, enabling larger training/inference and improving latency.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 12, 302-309
Coarse-Grained Access for 3-D-Stacked HBM-on-GPU Using DTCO of Fine-Pitch TSV and Integration Constraints
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
- Vol. 12, 302-309
Open AccessCoarse-Grained Access for 3-D-Stacked HBM-on-GPU Using DTCO of Fine-Pitch TSV and Integration Constraints
- Open Access
Capacitively-Coupled ECG and Respiration for Sleep–Wake Prediction and Risk Detection in Sleep Apnea Patients
Obstructive sleep apnea (OSA) patients would strongly benefit from comfortable home diagnosis, during which detection of wakefulness is essential. Therefore, capacitively-coupled electrocardiogram (ccECG) and bioimpedance (ccBioZ) sensors were used to record the sleep of suspected OSA patients, in parallel with polysomnography (PSG). The three objectives were quality assessment of the unobtrusive signals during sleep, prediction of sleep–wake using ccECG and ccBioZ, and detection of high-risk OSA patients. First, signal quality indicators (SQIs) determined the data coverage of ccECG and ccBioZ. Then, a multimodal convolutional neural network (CNN) for sleep–wake prediction was tested on these preprocessed ccECG and ccBioZ data. Finally, two indices derived from this prediction detected patients at risk. The data included 187 PSG recordings of suspected OSA patients, 36 (dataset “Test”) of which were recorded simultaneously with PSG, ccECG, and ccBioZ. As a result, two improvements were made compared to prior studies. First, the ccBioZ signal coverage increased significantly due to adaptation of the acquisition system. Secondly, the utility of the sleep–wake classifier increased as it became a unimodal network only requiring respiratory input. This was achieved by using data augmentation during training. Sleep–wake prediction on “Test” using PSG respiration resulted in a Cohen’s kappa () of 0.39 and using ccBioZ in = 0.23. The OSA risk model identified severe OSA patients with a of 0.61 for PSG respiration and of 0.39 using ccBioZ (accuracy of 80.6% and 69.4%, respectively). This study is one of the first to perform sleep–wake staging on capacitively-coupled respiratory signals in suspected OSA patients and to detect high risk OSA patients based on ccBioZ. The technology and the proposed framework could be applied in multi-night follow-up of OSA patients.Open AccessCapacitively-Coupled ECG and Respiration for Sleep–Wake Prediction and Risk Detection in Sleep Apnea Patients
Obstructive sleep apnea (OSA) patients would strongly benefit from comfortable home diagnosis, during which detection of wakefulness is essential. Therefore, capacitively-coupled electrocardiogram (ccECG) and bioimpedance (ccBioZ) sensors were used to record the sleep of suspected OSA patients, in parallel with polysomnography (PSG). The three objectives were quality assessment of the unobtrusive signals during sleep, prediction of sleep–wake using ccECG and ccBioZ, and detection of high-risk OSA patients. First, signal quality indicators (SQIs) determined the data coverage of ccECG and ccBioZ. Then, a multimodal convolutional neural network (CNN) for sleep–wake prediction was tested on these preprocessed ccECG and ccBioZ data. Finally, two indices derived from this prediction detected patients at risk. The data included 187 PSG recordings of suspected OSA patients, 36 (dataset “Test”) of which were recorded simultaneously with PSG, ccECG, and ccBioZ. As a result, two improvements were made compared to prior studies. First, the ccBioZ signal coverage increased significantly due to adaptation of the acquisition system. Secondly, the utility of the sleep–wake classifier increased as it became a unimodal network only requiring respiratory input. This was achieved by using data augmentation during training. Sleep–wake prediction on “Test” using PSG respiration resulted in a Cohen’s kappa () of 0.39 and using ccBioZ in = 0.23. The OSA risk model identified severe OSA patients with a of 0.61 for PSG respiration and of 0.39 using ccBioZ (accuracy of 80.6% and 69.4%, respectively). This study is one of the first to perform sleep–wake staging on capacitively-coupled respiratory signals in suspected OSA patients and to detect high risk OSA patients based on ccBioZ. The technology and the proposed framework could be applied in multi-night follow-up of OSA patients. - Open Access
- 24 September 2019
- Published by MDPI AG in Applied Sciences
- Vol. 9 (19) , 3994
Physiological Driver Monitoring Using Capacitively Coupled and Radar Sensors
Unobtrusive monitoring of drivers’ physiological parameters is a topic gaining interest, potentially allowing to improve the performance of safety systems to prevent accidents, as well as to improve the driver’s experience or provide health-related services. In this article, two unobtrusive sensing techniques are evaluated: capacitively coupled sensing of the electrocardiogram and respiration, and radar-based sensing of heartbeat and respiration. A challenge for use of these techniques in vehicles are the vibrations and other disturbances that occur in vehicles to which they are inherently more sensitive than contact-based sensors. In this work, optimized sensor architectures and signal processing techniques are proposed that significantly improve the robustness to artefacts. Experimental results, conducted under real driving conditions on public roads, demonstrate the feasibility of the proposed approach. R peak sensitivities and positive predictivities higher than 98% both in highway and city traffic, heart rate mean absolute error of 1.02 bpm resp. 2.06 bpm in highway and city traffic and individual beat R-R interval 95% percentile error within ±27.3 ms are demonstrated. The radar experimental results show that respiration can be measured while driving and heartbeat can be recovered from vibration noise using an accelerometer-based motion reduction algorithm.
- 24 September 2019
- Published by MDPI AG in Applied Sciences
- Vol. 9 (19) , 3994
Open AccessPhysiological Driver Monitoring Using Capacitively Coupled and Radar Sensors
Unobtrusive monitoring of drivers’ physiological parameters is a topic gaining interest, potentially allowing to improve the performance of safety systems to prevent accidents, as well as to improve the driver’s experience or provide health-related services. In this article, two unobtrusive sensing techniques are evaluated: capacitively coupled sensing of the electrocardiogram and respiration, and radar-based sensing of heartbeat and respiration. A challenge for use of these techniques in vehicles are the vibrations and other disturbances that occur in vehicles to which they are inherently more sensitive than contact-based sensors. In this work, optimized sensor architectures and signal processing techniques are proposed that significantly improve the robustness to artefacts. Experimental results, conducted under real driving conditions on public roads, demonstrate the feasibility of the proposed approach. R peak sensitivities and positive predictivities higher than 98% both in highway and city traffic, heart rate mean absolute error of 1.02 bpm resp. 2.06 bpm in highway and city traffic and individual beat R-R interval 95% percentile error within ±27.3 ms are demonstrated. The radar experimental results show that respiration can be measured while driving and heartbeat can be recovered from vibration noise using an accelerometer-based motion reduction algorithm.
- Open Access
- 1 September 2017
- Published by Walter de Gruyter GmbH in Current Directions in Biomedical Engineering
- Vol. 3 (2) , 301-304
Model based prediction of plateau pressure in mechanically ventilated patients
The risk of ventilator induced lung injury in mechanically ventilated (MV) critically ill patients can be mitigated by patient-specific optimisation of ventilator settings. Recent studies have shown that driving pressure, i.e. the difference between plateau pressure (Pplat) and PEEP, is a strong indicator for survival in MV patients suffering from ARDS. However, to measure Pplat, an extended end-inspiratory pause (EIP) has to be applied, possibly interrupting ventilation therapy. This study presents a method for predicting Pplat from normal breaths in MV patients.A total of 859 MV breaths with a 5 second EIP were recorded in 27 MV patients with ARDS. Two methods for determining Pplat were tested, one using an exponential fit of the pressure data and the other using a four-parameter viscoelastic model (VEM). Each method was identified using various lengths of data after the peak inspiratory pressure (PIP). Using the identified parameters, both methods were then used to predict the Pplat recorded at 5 seconds.The exponential method showed a median coefficient of variation (CV) from the real Pplat of 42.9% using data from PIP to 0.5 seconds after PIP, 24.9% using 1 second of data and 15.2% using 1.5 seconds of data. The respective VEM prediction median CVs were of 17.2%, 9.7% and 8.4%. Therefore, the VEM showed a better prediction than the non-physiological exponential model, allowing it to be used to reduce the clinical burden of determining Pplat by reducing the required length of the EIP to 1.5 seconds.
- 1 September 2017
- Published by Walter de Gruyter GmbH in Current Directions in Biomedical Engineering
- Vol. 3 (2) , 301-304
Open AccessModel based prediction of plateau pressure in mechanically ventilated patients
The risk of ventilator induced lung injury in mechanically ventilated (MV) critically ill patients can be mitigated by patient-specific optimisation of ventilator settings. Recent studies have shown that driving pressure, i.e. the difference between plateau pressure (Pplat) and PEEP, is a strong indicator for survival in MV patients suffering from ARDS. However, to measure Pplat, an extended end-inspiratory pause (EIP) has to be applied, possibly interrupting ventilation therapy. This study presents a method for predicting Pplat from normal breaths in MV patients.A total of 859 MV breaths with a 5 second EIP were recorded in 27 MV patients with ARDS. Two methods for determining Pplat were tested, one using an exponential fit of the pressure data and the other using a four-parameter viscoelastic model (VEM). Each method was identified using various lengths of data after the peak inspiratory pressure (PIP). Using the identified parameters, both methods were then used to predict the Pplat recorded at 5 seconds.The exponential method showed a median coefficient of variation (CV) from the real Pplat of 42.9% using data from PIP to 0.5 seconds after PIP, 24.9% using 1 second of data and 15.2% using 1.5 seconds of data. The respective VEM prediction median CVs were of 17.2%, 9.7% and 8.4%. Therefore, the VEM showed a better prediction than the non-physiological exponential model, allowing it to be used to reduce the clinical burden of determining Pplat by reducing the required length of the EIP to 1.5 seconds.