Search

In-memory computing for Machine Learning (ML) applications remedies the von Neumann bottlenecks by organizing computation to exploit parallelism and locality. Non-volatile memory devices such as Resistive RAM (ReRAM) offer integrated switching and storage capabilities showing promising performance for ML applications. However, ReRAM devices have design challenges, such as non-linear digital-analog conversion and circuit overheads. This paper proposes an In-Memory Boolean-to-Current Inference Architecture (IMBUE) that uses ReRAM-transistor cells to eliminate the need for such conversions. IMBUE processes Boolean feature inputs expressed as digital voltages and generates parallel current paths based on resistive memory states. The proportional column current is then translated back to the Boolean domain for further digital processing. The IMBUE architecture is inspired by the Tsetlin Machine (TM), an emerging ML algorithm based on intrinsically Boolean logic. The IMBUE architecture demonstrates significant performance improvements over binarized convolutional neural networks and digital TM in-memory implementations, achieving up to a 12.99x and 5.28x increase, respectively., Comment: Accepted at ACM/IEEE International Symposium on Low Power Electronics and Design 2023 (ISLPED 2023)

Heterogeneous multiprocessor systems have increased in complexity to provide both high performance and energy efficiency for a diverse range of applications. This motivates the need for a standard framework that enables the management, at runtime, of software applications executing on these processors. This paper proposes the first fully application- and platform-agnostic framework for runtime management approaches that control and optimise software applications and hardware resources. This is achieved by separating the system into three distinct layers connected by an API and cross-layer constructs called knobs and monitors. The proposed framework also supports the management of applications that are executing concurrently on heterogeneous platforms. The operation of the proposed framework is experimentally validated using a basic runtime controller and two heterogeneous platforms, to show how it is application- and platform-agnostic and easy to use. Furthermore, the management of concurrently executing applications through the framework is demonstrated. Finally, two recently reported runtime management approaches are implemented to demonstrate how the framework enables their operation and comparison. The energy and latency overheads introduced by the framework have been quantified and an open-source implementation has been released.

Wireless Sensor Networks (WSNs) offer a new solution for distributed monitoring, processing and communication. First of all, the stringent energy constraints to which sensing nodes are typically subjected. WSNs are often battery powered and placed where it is not possible to recharge or replace batteries. Energy can be harvested from the external environment but it is a limited resource that must be used efficiently. Energy efficiency is a key requirement for a credible WSNs design. From the power source's perspective, aggressive energy management techniques remain the most effective way to prolong the lifetime of a WSN. A new adaptive algorithm will be presented, which minimizes the consumption of wireless sensor nodes in sleep mode, when the power source has to be regulated using DC-DC converters. Another important aspect addressed is the time synchronisation in WSNs. WSNs are used for real-world applications where physical time plays an important role. An innovative low-overhead synchronisation approach will be presented, based on a Temperature Compensation Algorithm (TCA). The last aspect addressed is related to self-powered WSNs with Energy Harvesting (EH) solutions. Wireless sensor nodes with EH require some form of energy storage, which enables systems to continue operating during periods of insufficient environmental energy. However, the size of the energy storage strongly restricts the use of WSNs with EH in real-world applications. A new approach will be presented, which enables computation to be sustained during intermittent power supply. The discussed approaches will be used for real-world WSN applications. The first presented scenario is related to the experience gathered during an European Project (3ENCULT Project), regarding the design and implementation of an innovative network for monitoring heritage buildings. The second scenario is related to the experience with Telecom Italia, regarding the design of smart energy meters for monitoring the usage of household's appliances.

Monitoring current and voltage waveforms is fundamental to assess the power consumption of a system and to improve its energy efficiency. In this paper we present a smart meter for power consumption which does not need any electrical contact with the load or its conductors, and which can measure both current and voltage. Power metering becomes easier and safer and it is also self-sustainable because an energy harvesting module based on inductive coupling powers the entire device from the output of the current sensor. A low cost 32-bit wireless CPU architecture is used for data filtering and processing, while a wireless transceiver sends data via the IEEE 802.15.4 standard. We describe in detail the innovative contact-less voltage measurement system, which is based on capacitive coupling and on an algorithm that exploits two pre-processing channels. The system self-calibrates to perform precise measurements regardless the cable type. Experimental results demonstrate accuracy in comparison with commercial high-cost instruments, showing negligible deviations.