Driving Architectural Innovation for New Classes of Applications

Portrait of Matt Horsnell

Matt Horsnell, Senior Principal Research Engineer and Architecture Research Lead

Our architecture research teams investigate new applications, architectures, and associated technologies, while focusing on real-life use cases for Arm and its ecosystem. We work across all classes of architecture, microarchitecture, memory, and systems, to explore areas of improvement and new methods that can make the future of Arm architecture as ubiquitous as possible.

 

More broadly, we also recognize the importance of contributing to best practices in a continually evolving industry. In 2020, our paper, Rebasing Instruction Prefetching: An Industry Perspective, provided a new reference baseline for instruction prefetching research to bridge the gap between academia and industry. It was selected as one of IEEE Computer Architecture Letters’ top letters of the year and will hopefully pave the way to increasingly industry-relevant academic research breakthroughs.

We continue to demonstrate our commitment to the open-source community, which is an essential part of Arm’s ecosystem. We upstreamed support in gem5 for Arm's Transactional Memory Extension (TME), part of Arm’s A-profile Future Architecture Technologies program. This enables our partners and researchers to enhance their testing, benchmarking, and analysis capabilities as well as explore how this technology could evolve further.

 

Energy-driven compute, fueling the proliferation of tiny and low-cost IoT devices, continues to gain momentum, however, as existing modelling tools assume power is always available, these models aren’t suitable for the design and testing of such devices. To address this challenge, we developed Fused, an open-source full-system simulator which models execution, power consumption, and power supply in a closed loop. This means the interactions between them are correctly modelled, therefore enabling the hardware-software codesign and design space exploration that is crucial to the success of this technology.

Read More

Each year, we’re looking to find where we can get the next performance gains and as improvements within single nodes become more difficult to eke out, accelerated compute is climbing to the top of the agenda. As architecture researchers, it’s important for us to look at efficient integration, so we are working toward a deeper understanding of how applications and software can map onto increasingly complex systems that are made up of accelerators, CPUs, GPUs, and NPUs.

 

In addition, security is always a key concern, and so we are thinking about future architecture capabilities to mitigate against potential attack vectors on the architecture. We have to address this across the full spectrum – from A-profile, for example through our work on the CHERI and Morello projects, and how that maps all the way down into constrained embedded processors, such as in DARPA’s SSITH program. This work is crucial in order to better describe the security properties necessary in future software.

Certain elements of this year, such as forced remote working, has undeniably influenced the direction of technology research. I believe 2021 could be the year where extended reality (XR) could really take off as people are thinking more seriously about technologies where you can ‘feel’ presence even in the absence of colocation. The main challenge is that we’re still a couple of orders of magnitude away from the performance and efficiency that we need to reach the holistic experience, for example glasses with a virtual overlay, but I think devices working toward that are closer than ever. Researchers have an important role to play here. For example, at the SRC ADA Center as part of the ILLIXR project, they have been building out a complete full-system software stack to target XR. My hope is that this enables more academics to look at this problem, which can only push the field forward.