Architectural opportunities for novel dynamic EMI shifting (DEMIS)

Abstract

Processors emit non-trivial amounts of electromagnetic radiation, creating interference in frequency bands used by wireless communication technologies such as cellular, WiFi and Bluetooth. We introduce the problem of in-band radio frequency noise as a form of electromagnetic interference (EMI) to the computer architecture community as a technical challenge to be addressed. This paper proposes the new idea of Dynamic EMI Shifting (DEMIS) where architectural and/or compiler changes allow the EMI to be shifted at runtime. DEMIS processors dynamically move the interference from bands used during communication to other unused frequencies. Unlike previous works that leverage static techniques, DEMIS dynamically targets specific frequency bands; the type of techniques used here are only possible from an architectural perspective. This paper is also the first to provide insights in the new area of dynamic EMI shifting by evaluating several platforms and showing the EMI is sensitive to many architectural and compilation parameters. Our evaluation over real systems shows a decrease of in-band EMI ranging from 3 to 15 dB with less than a 10% average performance impact. A 15dB EMI reduction for LTE can represent over 3x bandwidth improvement for EMI bound communication. CCS CONCEPTS • Hardware → Noise reduction; Wireless devices; Signal integrity and noise analysis;

EMI Architectural Model and Core Hopping

“…Resets are excluded since they don't have periodic activity during the execution of a program and clocks are excluded because they are modulated using spread-spectrum techniques, which mean that clock cycles are slightly different from one another, and thus there is no single frequency for clock. This is confirmed by our experiments and observed in prior work [9].…”

“…Basically, this means that at every period of time, the data in each wire of interest will be collected for a certain number of cycles. As shown in prior work [5,9], the EMI changes as processors execute different phases of a process. For the purposes of this work, determining which execution phase is being processed is sufficient.…”

“…The most closely related work was published by Gorman et al introducing the impact EMI has on in-band wireless communications [9]. The authors show that small changes in architectural and code parameters can significantly impact the EMI emitted by a processor.…”

“…1 Micro-architects are uniquely positioned to solve the problem of EMI in a dynamic fashion, either through higher-level techniques, like code manipulations that will change activity rates on the SoC, or though low-level circuit techniques that change the wires emitting EMI. Recently, a dynamic in-band EMI reduction technique has been proposed [9], however, it requires measuring EMI during runtime and can only address EMI in a reactive way. This is mainly due to a distinct lack of tools or methodology for modeling and ultimately controlling EMI.…”

“…Thus, any model for EMI must be flexible enough to account for different processes. Therefore, previous works suggest simply monitoring the EMI for a runtime solution rather than attempting to create a model [9]. While, simply relying on run-time measurements is applicable in some use-cases, there is need for a design time technique that allows for estimating EMI, even if at lower accuracy.…”