Thermal Tracking and Estimation for Microprocessors

Due to increasing integration density and operating frequency of today's high performance processors, the temperature of a typical chip can easily exceed 100 degrees Celsius. However, the runtime thermal state of a chip is very hard to predict and manage due to the random nature in computing workloads, as well as the process, voltage and ambient temperature variability (together called PVT variability). The uneven nature (both in time and space) of the heat dissipation of the chip could lead to severe reliability issues and error-prone chip behavior (e.g. timing errors). Many dynamic power/thermal management techniques have been proposed to address this issue such as dynamic voltage and frequency scaling (DVFS), clock gating and etc. However, most of such techniques require accurate knowledge of the runtime thermal state of the chip to make efficient and effective control decisions. In this work we address the problem of tracking and managing the temperature of microprocessors which include the following sub-problems: (1) how to design an efficient sensor-based thermal tracking system on a given design that could provide accurate real-time temperature feedback; (2) what statistical techniques could be used to estimate the full-chip thermal profile based on very limited (and possibly noise-corrupted) sensor observations; (3) how do we adapt to changes in the underlying system's behavior, since such changes could impact the accuracy of our thermal estimation. The thermal tracking methodology proposed in this work is enabled by on-chip sensors which are already implemented in many modern processors. We first investigate the underlying relationship between heat distribution and power consumption, then we introduce an accurate thermal model for the chip system. Based on this model, we characterize the temperature correlation that exists among different chip modules and explore statistical approaches (such as those based on Kalman filter) that could utilize such correlation to estimate the accurate chip-level thermal profiles in real time. Such estimation is performed based on limited sensor information because sensors are usually resource constrained and noise-corrupted. We also took a further step to extend the standard Kalman filter approach to account for (1) nonlinear effects such as leakage-temperature interdependency and (2) varying statistical characteristics in the underlying system model. The proposed thermal tracking infrastructure and estimation algorithms could consistently generate accurate thermal estimates even when the system is switching among workloads that have very distinct characteristics. Through experiments, our approaches have demonstrated promising results with much higher accuracy compared to existing approaches. Such results can be used to ensure thermal reliability and improve the effectiveness of dynamic thermal management techniques.

Analysis of the "Sonar Hopf" Cochlea

The "Sonar Hopf" cochlea is a recently much advertised engineering design of an auditory sensor. We analyze this approach based on a recent description by its inventors Hamilton, Tapson, Rapson, Jin, and van Schaik, in which they exhibit the "Sonar Hopf" model, its analysis and the corresponding hardware in detail. We identify problems in the theoretical formulation of the model and critically examine the claimed coherence between the described model, the measurements from the implemented hardware, and biological data.

Control of bistability in a directly modulated semiconductor laser using delayed optoelectronic feedback

We show numerically that direct delayed optoelectronic feedback can suppress hysteresis and bistability in a directly modulated semiconductor laser. The simulation of a laser with feedback is performed for a considerable range of feedback strengths and delays and the corresponding values for the areas of the hysteresis loops are calculated. It is shown that the hysteresis loop completely vanishes for certain combinations of these parameters. The regimes for the disappearance of bistability are classified globally. Different dynamical states of the laser are characterized using bifurcation diagrams and time series plots.

Time domain passivity control for delayed teleoperation

La telepesencia combina diferentes modalidades sensoriales, incluyendo, entre otras, la visual y la del tacto, para producir una sensación de presencia remota en el operador. Un elemento clave en la implementación de sistemas de telepresencia para permitir una telemanipulación del entorno remoto es el retorno de fuerza. Durante una telemanipulación, la energía mecánica es transferida entre el operador humano y el entorno remoto. En general, la energía es una propiedad de los objetos físicos, fundamental en su mutual interacción. En esta interacción, la energía se puede transmitir entre los objetos, puede cambiar de forma pero no puede crearse ni destruirse. En esta tesis, se aplica este principio fundamental para derivar un nuevo método de control bilateral que permite el diseño de sistemas de teleoperación estables para cualquier arquitectura concebible. El razonamiento parte del hecho de que la energía mecánica insertada por el operador humano en el sistema debe transferirse hacia el entorno remoto y viceversa. Tal como se verá, el uso de la energía como variable de control permite un tratamiento más general del sistema que el control convencional basado en variables específicas del sistema. Mediante el concepto de Red de Potencia de Retardo Temporal (RPRT), el problema de definir los flujos de energía en un sistema de teleoperación es solucionado con independencia de la arquitectura de comunicación. Como se verá, los retardos temporales son la principal causa de generación de energía virtual. Este hecho se observa con retardos a partir de 1 milisegundo. Esta energía virtual es añadida al sistema de forma intrínseca y representa la causa principal de inestabilidad. Se demuestra que las RPRTs son transportadoras de la energía deseada intercambiada entre maestro y esclavo pero a la vez generadoras de energía virtual debido al retardo temporal. Una vez estas redes son identificadas, el método de Control de Pasividad en el Dominio Temporal para RPRTs se propone como mecanismo de control para asegurar la pasividad del sistema, y as__ la estabilidad. El método se basa en el simple hecho de que esta energía virtual debido al retardo debe transformarse en disipación. As__ el sistema se aproxima al sistema deseado, donde solo la energía insertada desde un extremo es transferida hacia el otro. El sistema resultante presenta dos cualidades: por un lado la estabilidad del sistema queda garantizada con independencia de la arquitectura del sistema y del canal de comunicación; por el otro, el rendimiento es maximizado en términos de fidelidad de transmisión energética. Los métodos propuestos se sustentan con sistemas experimentales con diferentes arquitecturas de control y retardos entre 2 y 900 ms. La tesis concluye con un experimento que incluye una comunicación espacial basada en el satélite geoestacionario ASTRA. ABSTRACT Telepresence combines different sensorial modalities, including vision and touch, to produce a feeling of being present in a remote location. The key element to successfully implement a telepresence system and thus to allow telemanipulation of a remote environment is force feedback. In a telemanipulation, mechanical energy must convey from the human operator to the manipulated object found in the remote environment. In general, energy is a property of all physical objects, fundamental to their mutual interactions in which the energy can be transferred among the objects and can change form but cannot be created or destroyed. In this thesis, we exploit this fundamental principle to derive a novel bilateral control mechanism that allows designing stable teleoperation systems with any conceivable communication architecture. The rationale starts from the fact that the mechanical energy injected by a human operator into the system must be conveyed to the remote environment and Vice Versa. As will be seen, setting energy as the control variable allows a more general treatment of the controlled system in contrast to the more conventional control of specific systems variables. Through the Time Delay Power Network (TDPN) concept, the issue of defining the energy flows involved in a teleoperation system is solved with independence of the communication architecture. In particular, communication time delays are found to be a source of virtual energy. This fact is observed with delays starting from 1 millisecond. Since this energy is added, the resulting teleoperation system can be non-passive and thus become unstable. The Time Delay Power Networks are found to be carriers of the desired exchanged energy but also generators of virtual energy due to the time delay. Once these networks are identified, the Time Domain Passivity Control approach for TDPNs is proposed as a control mechanism to ensure system passivity and therefore, system stability. The proposed method is based on the simple fact that this intrinsically added energy due to the communication must be transformed into dissipation. Then the system becomes closer to the ambitioned one, where only the energy injected from one end of the system is conveyed to the other one. The resulting system presents two benefits: On one hand, system stability is guaranteed through passivity independently from the chosen control architecture and communication channel; on the other, performance is maximized in terms of energy transfer faithfulness. The proposed methods are sustained with a set of experimental implementations using different control architectures and communication delays ranging from 2 to 900 milliseconds. An experiment that includes a communication Space link based on the geostationary satellite ASTRA concludes this thesis.

Some remarks on static-feedback linearization for time-varying systems

This work summarizes some results about static state feedback linearization for time-varying systems. Three different necessary and sufficient conditions are stated in this paper. The first condition is the one by [Sluis, W. M. (1993). A necessary condition for dynamic feedback linearization. Systems & Control Letters, 21, 277-283]. The second and the third are the generalizations of known results due respectively to [Aranda-Bricaire, E., Moog, C. H., Pomet, J. B. (1995). A linear algebraic framework for dynamic feedback linearization. IEEE Transactions on Automatic Control, 40, 127-132] and to [Jakubczyk, B., Respondek, W. (1980). On linearization of control systems. Bulletin del` Academie Polonaise des Sciences. Serie des Sciences Mathematiques, 28, 517-522]. The proofs of the second and third conditions are established by showing the equivalence between these three conditions. The results are re-stated in the infinite dimensional geometric approach of [Fliess, M., Levine J., Martin, P., Rouchon, P. (1999). A Lie-Backlund approach to equivalence and flatness of nonlinear systems. IEEE Transactions on Automatic Control, 44(5), 922-937]. (C) 2008 Elsevier Ltd. All rights reserved.

A feedback-based decentralised coordination model for distributed open real-time systems

Moving towards autonomous operation and management of increasingly complex open distributed real-time systems poses very significant challenges. This is particularly true when reaction to events must be done in a timely and predictable manner while guaranteeing Quality of Service (QoS) constraints imposed by users, the environment, or applications. In these scenarios, the system should be able to maintain a global feasible QoS level while allowing individual nodes to autonomously adapt under different constraints of resource availability and input quality. This paper shows how decentralised coordination of a group of autonomous interdependent nodes can emerge with little communication, based on the robust self-organising principles of feedback. Positive feedback is used to reinforce the selection of the new desired global service solution, while negative feedback discourages nodes to act in a greedy fashion as this adversely impacts on the provided service levels at neighbouring nodes. The proposed protocol is general enough to be used in a wide range of scenarios characterised by a high degree of openness and dynamism where coordination tasks need to be time dependent. As the reported results demonstrate, it requires less messages to be exchanged and it is faster to achieve a globally acceptable near-optimal solution than other available approaches.

Turn-on-time statistics of modulated lasers subjected to resonant weak optical feedback

We present analytical calculations of the turn-on-time probability distribution of intensity-modulated lasers under resonant weak optical feedback. Under resonant conditions, the external cavity round-trip time is taken to be equal to the modulation period. The probability distribution of the solitary laser results are modified to give reduced values of the mean turn-on-time and its variance. Numerical simulations have been carried out showing good agreement with the analytical results.

Movement time for motion-impaired users assisted by force-feedback: effects of movement amplitude, target width and gravity well width

This paper presents a study investigating how the performance of motion-impaired computer users in point and click tasks varies with target distance (A), target width (W), and force-feedback gravity well width (GWW). Six motion-impaired users performed point and click tasks across a range of values for A, W, and GWW. Times were observed to increase with A, and to decrease with W. Times also improved with GWW, and, with the addition of a gravity well, a greater improvement was observed for smaller targets than for bigger ones. It was found that Fitts Law gave a good description of behaviour for each value of GWW, and that gravity wells reduced the effect of task difficulty on performance. A model based on Fitts Law is proposed, which incorporates the effect of GWW on movement time. The model accounts for 88.8% of the variance in the observed data.

Real-time application of a constrained predictive controller based on dynamic neural networks with feedback linearization

This paper describes an experimental application of constrained predictive control and feedback linearisation based on dynamic neural networks. It also verifies experimentally a method for handling input constraints, which are transformed by the feedback linearisation mappings. A performance comparison with a PID controller is also provided. The experimental system consists of a laboratory based single link manipulator arm, which is controlled in real time using MATLAB/SIMULINK together with data acquisition equipment.

From saccades to smooth pursuit: real-time gaze control using motion feedback

The authors present an active vision system which performs a surveillance task in everyday dynamic scenes. The system is based around simple, rapid motion processors and a control strategy which uses both position and velocity information. The surveillance task is defined in terms of two separate behavioral subsystems, saccade and smooth pursuit, which are demonstrated individually on the system. It is shown how these and other elementary responses to 2D motion can be built up into behavior sequences, and how judicious close cooperation between vision and control results in smooth transitions between the behaviors. These ideas are demonstrated by an implementation of a saccade to smooth pursuit surveillance system on a high-performance robotic hand/eye platform.

Metastable Periodic Patterns in Singularly Perturbed Delayed Equations

We consider the scalar delayed differential equation epsilon(x) over dot(t) = -x(t) + f(x(t-1)), where epsilon > 0 and f verifies either df/dx > 0 or df/dx < 0 and some other conditions. We present theorems indicating that a generic initial condition with sign changes generates a solution with a transient time of order exp(c/epsilon), for some c > 0. We call it a metastable solution. During this transient a finite time span of the solution looks like that of a periodic function. It is remarkable that if df/dx > 0 then f must be odd or present some other very special symmetry in order to support metastable solutions, while this condition is absent in the case df/dx < 0. Explicit epsilon-asymptotics for the motion of zeroes of a solution and for the transient time regime are presented.

Dynamic Tracking with Zero Variation and Disturbance Rejection Applied to Discrete-Time Systems

Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)