Tutorials

Participants will engage deeply with quantum machine learning (QML) and quantum architecture search (QAS) through interactive, hands-on sessions leveraging open-source quantum computer simulator such as Qiskit and PennyLane, with a specific focus on both computational intelligence applications and quantum circuit design. This tutorial provides a comprehensive learning experience, starting with foundational concepts in quantum information science (QIS), such as qubits and quantum gates, and advancing to key methodologies in QML and quantum circuit optimization. Attendees will explore a variety of QML models, including quantum neural networks (QNN), quantum convolutional neural networks (QCNN) and quantum recurrent neural networks (QRNN), with a focus on their application to AI/ML. In parallel, participants will dive into QAS techniques, which focus on discovering and optimizing quantum circuit architectures for these models, improving both performance and resource efficiency. 

This tutorial aims to present the most recent advancements in the field of systematic analog and RF IC design. The goal is to enrich designers' insights and enhance their eBiciency in the algorithmic design of analog circuits. Moreover, fundamental design and modeling algorithms will be exploited to develop design automation tools, a highly demanding topic in today's IC design industry.

Chiplets and 2.5D integration is an elegant solution for integrating a heterogeneous mix of components to reduce footprint, increase signal speed, and reduce power. Designing a 2.5D device is much faster and easier than designing a conventional system-on-chip (SoC) because it can use existing components. Modifying the design can be as simple as swapping out one component for another – an enormous benefit in IP reuse and in cost and risk reduction. In this three-part tutorial we will first introduce the audience to the technology foundations and industry standards for advanced 2.5D integration and packaging using chiplets [1,2]. The technology introduction is followed by two examples highlighting the system co-design using 2.5D and chiplets. The first is an advanced RF chiplet based active antenna using reconfigurable metasurfaces [3] which are man-made surfaces, which consist of sub-wavelength periodic elements—meta-atoms—that can be reconfigured to manipulate incoming electromagnetic waves. The second example is an architecture for real-time processing of data that originates in large format tiled imaging arrays used in wide area motion imagery ubiquitous surveillance [4].

The security and reliability of the Internet of Things (IoT) devices is critical as they are increasingly integrated into more complex and safety-critical systems, such as for example in healthcare for continuous monitoring and patient management, the Internet of Medical Things (IoMT). The importance of addressing these issues lies at the intersection of communication systems and control systems, which are essential for ensuring both the reliability of transmitted data and the control algorithms responsible for device functionality.

In recent years, machine learning has been extensively applied for the modeling and optimization of integrated circuits. While learning techniques are seamlessly added to improve existing digital synthesis flows, the development of learning-based analog EDA faces more challenges and lags behind the digital counterpart as analog design is usually performed in a customized and manual approach.

The objective of this tutorial is to provide an overview of recent progress in the application of machine learning to analog EDA. State-of-the-art learning and optimization techniques for the modeling and design of analog ICs are presented and discussed. Practical considerations, challenges, and opportunities of ML for analog EDA will be provided.

Direct Digital Frequency Synthesis (DDFS) was initially introduced by J. Tierney in 1971 [1]. This digital signal processing (DSP) technique represents a digital pattern in the analog domain based on a fixed clock period (frequency). Recently, DDFS has regained popularity due to its ability to exploit frequency/phase multiplexing/modulation for high-fidelity multi-Qubit controllers [2]. A digitally intensive DDFS-based controller with modulation capabilities can benefit from technology scaling, achieving fast and continuous phase switching. CMOS-based DDFS controllers have already demonstrated their effectiveness [3]. However, achieving low-power, high-speed DDFS remains challenging, especially for large-scale quantum computers operating at cryogenic temperatures (CT). Non-idealities caused by increased threshold voltage at CT, device mismatch, and PVT variations must be carefully addressed. 

The brain is the perfect place to look for inspiration to develop more efficient neural networks. While the computational cost of deep learning exceeds millions of dollars to train large-scale models, our brains are somehow equipped to process an abundance of signals from our sensory periphery within a power budget of approximately 10-20 watts. How can we enable energy efficient language processing within the same energy budget as the brain?

This two-part tutorial delivered by two experts in the field of implantable neurotechnologies will familiarize the audience with the fundamentals of brain interfacing (i.e., sensing and modulation) in less explored, emerging modalities involving neurochemical signals and ultrasound waves.

As AI scales, connectivity has proven to be a key bottleneck. Thus, AI software and hardware developers must be familiar with the relevant communication technologies and trends. Continued advances in CMOS, advanced packaging, and chiples for AI computing are increasing bandwidth demands over all parts of the network. Meanwhile, power is tightly constrained, and new requirements for ultra-low-latency links have emerged. This tutorial affords the reader with background on the wide variety of key connectivity technologies being brought to bear on these emergent challenges. This includes the challenges and opportunities for electrical cable and optical fibre links, alternative forms of modulation and error correction.

The promising future of fully immersive virtual reality (VR) platforms is heavily tied to the design and development of reliable and seamless human machine interfaces (HMIs). HMIs involve a hardware module that senses the intended activity and a software module that classifies it. This classifier can be readily implemented using machine learning algorithms but there are various options available for the sensing module. An intuitive option for a seamless interaction with the computer is to detect the biosignals, which are essentially by-products of a given human activity. The added value of this modality, for example in contrast to computer vision, is that it can be deployed using a collection of wearable devices for mobile solutions. For example, eye gaze direction can be detected using electrooculograms (EOG) and hand gestures can be detected using electromyograms (EMG) from the forearm. Apart from detecting interactions, other forms of biosignal can be utilised to convey other attributes associated with the user (e.g., electrocardiogram (ECG) or electroencephalogram (EEG)). Beyond VR, biosignal-based HMIs have a wider range of applications in various assistive technologies. 

Nowadays, there is growing trend toward the design of sustainable circuits and systems, in which DC-power consumption needs to be reduced. RF isolators are commonly exploited in RF front-ends chains to prevent active stages to be damaged by the presence of RF signal-power reflections created by the passive stages (e.g., pre-select RF bandpass filters) in their out-of-band regions. Indeed, such RF signal-power reflections may induce RF amplifiers to operate in non-linear regime or to create additional undesired mixing products in frequency-conversion stages, which may result in the malfunctioning of the entire RF transceiver. However, this is done at the expense of higher size/volume and increased DC power consumption, respectively, as well as higher cost. A more-efficient solution that is gaining a huge research attention may be the exploitation of “absorptive” or “reflectionless” RF devices, where the out-of-band signal energy is dissipated inside the circuit instead of being reflected back to the source so that RF isolators are no longer needed.