Browse Prior Art Database

Method to Reduce Chip Hot Spots Using Redundant Macros and Units

IP.com Disclosure Number: IPCOM000030275D
Original Publication Date: 2004-Aug-04
Included in the Prior Art Database: 2004-Aug-04
Document File: 2 page(s) / 34K

Publishing Venue

IBM

Abstract

Proposed is the use of redundant functional units and/or circuit macros to reduce thermal hot spots that are caused by the execution of tight computational loops in microprocessors. Reducing such hot spots can reduce thermal cooling and packaging costs while potentially increasing performance.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 1 of 2

Method to Reduce Chip Hot Spots Using Redundant Macros and Units

The power consumption and associated heat dissipation of CMOS circuits used in modern microprocessor manufacturing can be broken down into static and dynamic components. Static power dissipation is the portion of power consumption that is not related to activity, or switching, of transistors and includes leakage current. The idea presented here is related to active or dynamic power dissipation that is related to the voltage and frequency at which circuits are exercised. Power-aware microprocessors can reduce some of the static power consumed by idle circuits by employing a variety of circuit techniques. Unfortunately, when active circuits are heavily utilized, their power consumption cannot be reduced since it is a function of the technology used to implement them as well as their operating voltage and frequency as seen in the following equation:

2Pactive= CeffVf

C is the effective capacitance of the circuit in question, V is its operating voltage, and f is its frequency. One of the primary design constraints for future microprocessors is the management of thermal dissipation due to high power consumption. Thermal heat dissipation and cooling affects the overall cost and reliability of microprocessors; therefore, techniques that can reduce cooling and packaging requirements have a positive impact on future designs. Since the power consumption of various microprocessor units varies with transistor density and utilization, some areas of a die may tend to operate at considerably higher temperatures than others. Such areas are commonly referred to as hot spots. Hot spots result in two significant design limitations. First, since silicon semiconductor performance degrades with temperature increases, the existence of hot spots limits the maximum operating frequency at which a processor can reliably operate. Second, to maintain hot spot areas within the desirable operating range, it is often necessary to design heat spreading and dissipation hardware that is more costly since it is over designed for most areas of the die.

Redudant Macros and Units

Functional units, such as integer or floating-point execution units, can cause hot spots in some microprocessor designs when they are highly utilized. Such conditions can arise in tight computational loops that do not contain long dependence chains or experience good cache behavior. One technique to handle such hot spots is to reduce or throttle instruction issue when thermal sensors detect temperatures exceeding certain thresholds [1]. This technique is effective, but comes at a possibly significant performance penalty. Another techni...