Browse Prior Art Database

Management of Dynamically Resizable Data Processing Units based on Application State Predictive Estimation.

IP.com Disclosure Number: IPCOM000196384D
Publication Date: 2010-Jun-02
Document File: 5 page(s) / 200K

Publishing Venue

The IP.com Prior Art Database

Abstract

Existing data processing units dynamically resizing algorithms/methods do not consider predictable behavior (future system states transitions) of an application. They can use additional instruction processing units but do not resize the CPU (ALU, instruction processing units (like DSP, floating point, etc), memories, caches and other processing blocks) bit-width depending on application state transitions and the current power source charge level. They do not put parts of CPU and its support logic which are not used in the current system bit-width mode in the low power states. All these listed functions are covered by the proposed module for management of dynamically resizable data processing units based on application state predictive estimation.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 29% of the total text.

Management of Dynamically Resizable Data Processing Units based on Application State Predictive Estimation.

Abstract:

Existing data processing units dynamically resizing algorithms/methods do not consider predictable behavior (future system states transitions) of an application. They can use additional instruction processing units but do not resize the CPU (ALU, instruction processing units (like DSP, floating point, etc), memories, caches and other processing blocks) bit-width depending on application state transitions and the current  power source charge level. They do not put parts of CPU and its support logic which are not used in the current system bit-width mode in the low power states. All these listed functions are covered by the proposed module for management of dynamically resizable data processing units based on application state predictive estimation.

Currently existing  dynamically resizable processing units (DRPU ) and their managers (DRPCM) very often have limitations in adjustment of  the processing unit size (8-,16-,32- or 64-bit mode instructions execution) up to fast changing application environment states and their power consumption requirements based on future application state prediction (FASP). Introduction of the DRPCM allows to minimize power consumption in run and low-power modes and to not exceed max power consumption level. The DRPCMs also protect systems/devices from unnecessary DRPU loads and from reaching critical power consumption levels caused by faulty processing requests and/or wrong system state transitions.

As recent mobile (battery powered) or automotive MCU systems have to dynamically align their compute resources up to the power consumption requirements in run and low-power modes of operation. Predictable management of this balance in safety oriented applications becomes really crucial for delivering required compute power while prolonging operation time between battery charges.

To profit from listed above advantages, a device was proposed which is capable of controlling unit dynamic resizing in a MCU system. It can select only a part of the MCU, its ALU and surrounding blocks (instruction modules, caches, etc) to process upcoming tasks. The selected slices of memories containing instructions and data can be partially/fully switched off to save power.  The DRPCM in the proposed module uses the FASP mechanism to align compute resources up to the power consumption requirements and to omit illegal state transitions.

The DRPCM processes incoming resourcing requests in two main steps. First, on incoming request it analyzes possible state transitions which can be caused by this request from the current application state. It also checks the request control counters if this request is not happening too often. In this first step the request will be rejected if the current application state does not allow any transition on this request or it has happened too often so that the counter related to this transitio...