Andrea Tilli

Thermal and Energy Management of High-Performance Multicores: Distributed and Self-Calibrating Model-Predictive Controller

As result of technology scaling, single-chip multicore power density increases and its spatial and temporal workload variation leads to temperature hot-spots, which may cause nonuniform ageing and accelerated chip failure. These critical issues can be tackled by closed-loop thermal and reliability management policies. Model predictive controllers (MPC) outperform classic feedback controllers since they are capable of minimizing performance loss while enforcing safe working temperature. Unfortunately, MPC controllers rely on a priori knowledge of thermal models and their complexity exponentially grows with the number of controlled cores. In this paper, we present a scalable, fully distributed, energy-aware thermal management solution for single-chip multicore platforms. The model-predictive controller complexity is drastically reduced by splitting it in a set of simpler interacting controllers, each one allocated to a core in the system. Locally, each node selects the optimal frequency to meet temperature constraints while minimizing the performance penalty and system energy. Comparable performance with state-of-the-art MPC controllers is achieved by letting controllers exchange a limited amount of information at runtime on a neighborhood basis. In addition, we address model uncertainty by supporting learning of the thermal model with a novel distributed self-calibration approach that matches well the controller architecture.

A distributed and self-calibrating model-predictive controller for energy and thermal management of high-performance multicores

High-end multicore processors are characterized by high power density with significant spatial and temporal variability. This leads to power and temperature hot-spots, which may cause non-uniform ageing and accelerated chip failure. These critical issues can be tackled on-line by closed-loop thermal and reliability management policies. Model predictive controllers (MPC) outperform classic feedback controllers since they are capable of minimizing a cost function while enforcing safe working temperature. Unfortunately basic MPC controllers rely on a-priori knowledge of multicore thermal model and their complexity exponentially grows with the number of controlled cores. In this paper we present a scalable, fully-distributed, energy-aware thermal management solution. The model-predictive controller complexity is drastically reduced by splitting it in a set of simpler interacting controllers, each allocated to a core in the system. Locally, each node selects the optimal frequency to meet temperature constraints while minimizing the performance penalty and system energy. Global optimality is achieved by letting controllers exchange a limited amount of information at run-time on a neighbourhood basis. We address model uncertainty by supporting learning of the thermal model with a novel distributed self-calibration approach that matches well the controller architecture.

SCC thermal model identification via advanced bias-compensated least-squares

Compact thermal models and modeling strategies are today a cornerstone for advanced power management to counteract the emerging thermal crisis for many-core systems-on-chip. System identification techniques allow to extract models directly from the target device thermal response. Unfortunately, standard Least Squares techniques cannot effectively cope with both model approximation and measurement noise typical of real systems. In this work, we present a novel distributed identification strategy capable of coping with real-life temperature sensor noise and effectively extracting a set of low-order predictive thermal models for the tiles of Intels Single-chip-Cloud-Computer (SCC) many-core prototype.

Don't burn your mobile!: safe computational re-sprinting via model predictive control

Computational sprinting has been recently proposed as an effective solution to mitigate the upcoming dark silicon utilization wall problems. As most applications do not require constantly the maximum performance level and parallelism, computational sprinting uses the intrinsic thermal capacitance of the heat dissipation system, possibly augmented with Phase Change Materials, as heat buffer for tolerating bursts of intensive computational resource usage largely exceeding the steady-state thermal design power. It is clear that sprinting poses peculiar challenges on the dynamic thermal control policy. In this paper we introduce and evaluate an innovative and low-overhead hierarchical model-predictive controller that enables thermally-safe sprinting while guaranteeing a predictable re-sprinting rate.

A synchronous coordinates approach in position and speed estimation for Permanent Magnet Synchronous Machines

A novel and simple observer scheme is proposed for rotor speed and position reconstruction for Permanent Magnet Synchronous Machines. The reference frame adopted in the observer is pushed toward the synchronous one by forcing it to be intrinsically aligned with the estimated back-emf vector and by designing suitable adaptations law for its speed and angle along with the back-emf amplitude. Stator flux dynamics are not used in this approach, leading to an improved robustness with respect to voltage and current measurement uncertainties. Stability analysis is carried out by using singular perturbation approach. Effective tuning rules are drawn exploiting insightful linearization of the proposed nonlinear adaptive observer. Simulations show the properties of the presented method.

Thermal models characterization for reliable temperature capping and performance optimization in Multiprocessor Systems on Chip

Modern Multiprocessor Systems-on-Chip (MP-SoC) offer high computing performance at the expense of huge power densities unevenly distributed on the chip. This generates hot spots that may cause performance and reliability degradations as well as power consumption increases. In recent years several thermal control strategies have been developed to avoid the occurrences of these hot spots. In particular, schemes based on Model Predictive Control (MPC) theory represent the actual state-of-the-art due to their capability to explicitly deal with constraints. In this paper we discuss some important properties for the design of predictive controllers with constraints for the class of thermal system. Starting from the general partial differential equation representing the heat diffusion in a solid, the feasibility and a useful property for the reduction of the number of constraints are proven. Moreover, exploiting theoretical results, a two layers control architecture is presented, which is capable of ensuring feasibility in every circumstance. Simulative results show the benefits of this approach.

Globally asymptotically stable reconstruction of flux, rotor position and speed for permanent magnets machines

An original full-order observer is presented for stator fluxes and speed reconstruction for Permanent Magnet Synchronous Machines. The estimation is carried out in a fixed reference frame exploiting the stator currents and voltages as the only known variables. Global asymptotic convergence properties of the proposed scheme are drawn, casting the estimation problem into adaptive systems theoretical framework. Insightful modification to the flux estimation dynamics is proposed in order to improve the observer performance in the low speed region. Simulation tests asses the features of the presented method.

Induction motor sensorless observer aligned with rotor flux derivative

A novel observer is proposed to reconstruct rotor speed and rotor flux vector in induction motors. The presented solution is characterized by a simple structure, based on a rotating reference frame which is pushed toward perfect alignment with the time-derivative of the rotor flux vector with respect to the stator windings reference frame. This approach allows to derive rotor speed and flux estimations without relying on pure integration of the stator flux dynamics and without using typical adaptive schemes which exploit the product between flux components and speed in the back-EMF of the stator current dynamics. Therefore, very good robustness is achieved with respect to voltage and current measurement uncertainties. Stability analysis is carried out by singular perturbation arguments, guaranteeing semiglobal practical asymptotic convergence. Effective tuning rules are derived considering a meaningful linearization of the proposed nonlinear observer. Simulations are presented to show the effectiveness of the solution.

Modeling and control design for power systems driven by battery/supercapacitor hybrid energy storage devices

This paper addresses several important problems in a typical power system driven by battery/supercapacitor hybrid energy storage devices. The currents in the battery and the supercapacitor are actively controlled by two bidirectional buck-boost converters. Detailed investigation is conducted on deriving two state-space averaged models for the whole interconnected system. The control objective is to track the references of two variables. The control design problem is converted into a numerically efficient optimization problem with linear matrix inequality (LMI) constraints. When the optimization algorithm is applied to an experimental system, the resulting optimal control law is very close to a simple integrator control (with two inputs and two outputs). The simple integrator control is applied to the system in both simulation and experiment. The results confirm the effectiveness of the modeling and control design methods.

Anti-windup scheme for current control of Shunt Active Filters

An anti-windup technique for current control of Shunt Active Filters is presented. The main purpose of the proposed solution is to preserve the unconstrained tracking error dynamics, even under saturated conditions. A suitable modification of current references is imposed through the design of an additional dynamics. The proposed approach is valid for any kind of controller adopted to steer SAF current; in this work it has been applied on an internal model based controller. A variant of the basic approach is proposed to improve performances. Simulation tests confirm the effectiveness of the presented solution.