Chip-level power and thermal implications will continue to rule as one of the primary design constraints and performance limiters. The gap between average and peak power actually widens with increased levels of core integration. As such, if per-core control of power levels (modes) is possible, a global power manager should be able to dynamically set the modes suitably. This would be done in tune with the workload characteristics, in order to always maintain a chip-level power that is below the specified budget. Furthermore, this should be possible without significant degradation of chip-level throughput performance. We analyze and validate this concept in detail in this paper. We assume a per-core DVFS (dynamic voltage and frequency scaling) knob to be available to such a conceptual global power manager. We evaluate several different policies for global multi-core power management. In this analysis, we consider various different objectives such as prioritization and optimized throughput. Overall, our results show that in the context of a workload comprised of SPEC benchmark threads, our best architected policies can come within 1% of the performance of an ideal oracle, while meeting a given chip-level power budget. Furthermore, we show that these global dynamic management policies perform significantly better than static management, even if static scheduling is given oracular knowledge.

/pdf/an-analysis-of-efficient-multi-core-global-power-management-1ni9trrk1b.pdf

An Analysis of Efficient Multi-Core Global Power Management Policies: Maximizing Performance for a Given Power Budget

Within-die process variation causes individual cores in a ChipMultiprocessor (CMP) to differ substantially in both static powerconsumed and maximum frequency supported. In this environment,ignoring variation effects whenscheduling applications or when managing power withDynamic Voltage and Frequency Scaling (DVFS) is suboptimal. This paper proposes variation-aware algorithms for applicationscheduling and power management. One such power managementalgorithm, called {\em LinOpt}, uses linear programmingto find the best voltage and frequency levels for each of thecores in the CMP --- maximizing throughput at a given power budget.In a 20-core CMP, the combination of variation-awareapplication scheduling and {\em LinOpt} increases the averagethroughput by 12--17\% and reduces the average $ED^2$ by 30--38\%--- all relative to using variation-awarescheduling together with a simple extension to Intel's Foxtonpower management algorithm.

/pdf/variation-aware-application-scheduling-and-power-management-29z1mmaf3m.pdf

Variation-Aware Application Scheduling and Power Management for Chip Multiprocessors

A 65 nm resilient circuit test-chip is implemented with timing-error detection and recovery circuits to eliminate the clock frequency guardband from dynamic supply voltage (V CC ) and temperature variations as well as to exploit path-activation probabilities for maximizing throughput. Two error-detection sequential (EDS) circuits are introduced to preserve the timing-error detection capability of previous EDS designs while lowering clock energy and removing datapath metastability. One EDS circuit is a dynamic transition detector with a time-borrowing datapath latch (TDTB). The other EDS circuit is a double-sampling static design with a time-borrowing datapath latch (DSTB). In comparison to previous EDS designs, TDTB and DSTB redirect the highly complex metastability problem from both the datapath and error path to only the error path, enabling a drastic simplification in managing metastability. From a survey of various EDS circuit options, TDTB represents the lowest clock energy EDS circuit known; DSTB represents the lowest clock energy static-EDS circuit with SER protection known. Error-recovery circuits are introduced to replay failing instructions at lower clock frequency to guarantee correct functionality. Relative to conventional circuits, test-chip measurements demonstrate that resilient circuits enable either 25%-32% throughput gain at equal V CC or at least 17% V CC reduction at equal throughput, corresponding to 31%-37% total power reduction.

Energy-Efficient and Metastability-Immune Resilient Circuits for Dynamic Variation Tolerance

A 65 nm resilient circuit test-chip is implemented with timing-error detection and recovery circuits to eliminate the clock frequency guardband from dynamic supply voltage (VCC) and temperature variations as well as to exploit path-activation probabilities for maximizing throughput. Two error-detection sequential (EDS) circuits are introduced to preserve the timing-error detection capability of previous EDS designs while lowering clock energy and removing datapath metastability. One EDS circuit is a dynamic transition detector with a time-borrowing datapath latch (TDTB). The other EDS circuit is a double-sampling static design with a time-borrowing datapath latch (DSTB). In comparison to previous EDS designs, TDTB and DSTB redirect the highly complex metastability problem from both the datapath and error path to only the error path, enabling a drastic simplification in managing metastability. From a survey of various EDS circuit options, TDTB represents the lowest clock energy EDS circuit known; DSTB represents the lowest clock energy static-EDS circuit with SER protection known. Error-recovery circuits are introduced to replay failing instructions at lower clock frequency to guarantee correct functionality. Relative to conventional circuits, test-chip measurements demonstrate that resilient circuits enable either 25%-32% throughput gain at equal VCC or at least 17% VCC reduction at equal throughput, corresponding to 31%-37% total power reduction.

This paper briefly reviews the forces that caused the power problem, the solutions that were applied, and what the solutions tell us about the problem As systems became more power constrained, optimizing the power became more critical; viewing power reduction from an optimization perspective provides valuable insights Section III describes these insights in more detail, including why Vdd and Vth have stopped scaling Section IV describes some of the low power techniques that have been used in the past in the context of the optimization framework This framework also makes it easy to see the impact of variability, which is discussed in more detail in section V along with the adaptive mechanisms that have been proposed and deployed to minimize the energy cost Section VI describes possible strategies for dealing with the slowdown in gate energy scaling, and the final section concludes by discussing the implications of these strategies for device designers

Scaling, power, and the future of CMOS

This paper describes the embedded feedback and control system on a 90-nm Itanium family processor, code-named Montecito, that maximizes performance while staying within a target power and temperature (PT) envelope. This system, referred to as Foxton Technology (FT), utilizes on-chip sensors and an embedded microcontroller to measure PT and modulate both voltage and frequency (VF) to optimize performance while meeting PT constraints. Changing both VF takes advantage of the cubic relationship of P/spl prop/CV/sup 2/F. We present measured results that show a 31% reduction in power for only a 10% drop in frequency. Montecito is able to implement FT using only 0.5% of the die area and 0.5% of the die power.

/pdf/power-and-temperature-control-on-a-90-nm-itanium-family-26sp1zp45l.pdf

Power and temperature control on a 90-nm Itanium family processor

In an embodiment, a processor comprises: a plurality of cores each to execute instructions; a plurality of thermal sensors, at least one of which is associated with each of the cores; and a power control unit (PCU) coupled to the cores. The PCU includes a thermal control logic to preemptively throttle a first core by a first throttle amount when a temperature of a second core exceeds at least one thermal threshold. Note that this first core may be preemptively throttled independently of a throttling of the second core and may have a temperature of the first core does not exceed any thermal threshold. Other embodiments are described and claimed.

Adaptive algorithm for thermal throttling of multi-core processors with non-homogeneous performance states

Methods and systems to detect droop events on-chip, which may include a sensor circuit located adjacent to a voltage node to convert a corresponding voltage to a digital count or value indicative of the voltage. The sensor circuit may include an n-stage ring oscillator and an asynchronous counter. The sensor circuit may include circuitry to capture and convert a phase associated with a count to a binary fractional value to increase voltage resolution. Multiple counts associated with the node may be evaluated at the node to identify minimum and maximum counts and corresponding time stamps. More complex evaluation and control circuitry may be shared amongst a plurality of sensor circuits and may include circuitry to generate and compare counts to one or more variable thresholds, circuitry to average counts over time, and memory to store state values associated with the sensors.

Methods and systems to detect voltage changes within integrated circuits

Disclosed herein are different embodiments for estimating and/or controlling power consumption in a chip based on hot and cool temperatures in the chip.

Power estimation for a semiconductor device

Structured, or tabular, data is the most common format in data science. While deep learning models have proven formidable in learning from unstructured data such as images or speech, they are less accurate than simpler approaches when learning from tabular data. In contrast, modern tree-based Machine Learning (ML) models shine in extracting relevant information from structured data. An essential requirement in data science is to reduce model inference latency in cases where, for example, models are used in a closed loop with simulation to accelerate scientific discovery. However, the hardware acceleration community has mostly focused on deep neural networks and largely ignored other forms of machine learning. Previous work has described the use of an analog content addressable memory (CAM) component for efficiently mapping random forests. In this work, we focus on an overall analog-digital architecture implementing a novel increased precision analog CAM and a programmable network on chip allowing the inference of state-of-the-art tree-based ML models, such as XGBoost and CatBoost. Results evaluated in a single chip at 16nm technology show 119x lower latency at 9740x higher throughput compared with a state-of-the-art GPU, with a 19W peak power consumption.

James S. Ignowski

Papers

Power and temperature control on a 90-nm Itanium family processor

Adaptive algorithm for thermal throttling of multi-core processors with non-homogeneous performance states

Methods and systems to detect voltage changes within integrated circuits

Power estimation for a semiconductor device

X-TIME: An in-memory engine for accelerating machine learning on tabular data with CAMs