METHOD AND APPARATUS FOR ON-DIE TEMPERATURE SENSING AND CONTROL
For one disclosed embodiment, a plurality of processor cores may be on a semiconductor die. The processor cores may have at least one corresponding temperature sensor. Circuitry on the semiconductor die may generate thermal event indications based on sensed temperatures from multiple temperature sensors of multiple processor cores. A thermal event indication may indicate that a sensed temperature exceeds a temperature point. Central management logic on the semiconductor die may receive thermal event indications based on sensed temperatures from multiple temperature sensors of multiple processor cores. The central management logic may modify operation of one or more of the processor cores in response to a thermal event indication. Other embodiments are also disclosed.
This patent application is a continuation application of U.S. application Ser. No. 11/025,140, filed Dec. 30, 2004.TECHNICAL FIELD
Embodiments of the present invention relate to devices and methods for temperature measurement of processor cores, and more particularly to on-die circuits and methods for temperature measurement and reporting.BACKGROUND
Modern central processing units (“CPUs”), with increasing processor core frequencies and power densities, are rapidly reaching a point in which the CPU performance is limited by the amount of heat that can be extracted from the CPU by cooling technology.
High performance CPUs sometimes integrate an on-die thermal diode that converts junction temperatures to some electric value. External analog-to-digital (“A/D”) devices can be found on several computer platforms, converting the electric value into usable information for temperature control.
Due to routing and accuracy limitations, the thermal diode is placed in locations which are not at a hotspot of the CPU. A hotspot is a localized region having a higher temperature than regions of the processor adjoining the hotspot. Temperature differences, as high as 15° C. have been observed between the diode location and the hotspot. The result is inaccurate temperature readings and discrepancy between the reported value and the actual temperature used for device specification and control. Inaccurate temperature reading leads to inclusion of higher design margins, increased cooling costs and limited functionality.
Another thermal protection mechanism uses a separate sensor to detect a maximum threshold temperature at the hottest hotspot on the die. Once the threshold is reached, a single trigger occurs and thermal protection is initiated. Other than this single trigger, the sensor provides no other feedback for use in thermal or acoustic management.
If this second sensor is external to the device, the external sensor suffers from time delay and accuracy problems resulting from the physical separation of the sensor from the hotspot.
There is a need for improved on-die temperature monitoring to improve performance in a thermally constrained environment. There is also a need for faster and more accurate reporting, eliminating gaps between thermal measurement and control.
Embodiments of the present invention integrate a sensor at a hotspot of the die for the measurement and reporting of temperature values, and for the generation of notifications. Using a single sensor to provide both temperature-responsive triggers and temperature measurement improves the speed, consistency, and accuracy of temperature reporting. Processor operation can be optimized, maximizing performance in a thermally constrained environment.
The temperature responsive triggers may be utilized to direct both on-die and off-die temperature control, including directing the processor to modify operation, activating external fans or cooling systems, modifying the output of an external power supply, and/or putting a portion or an entirety of the processor into a standby state. Examples of modifying processor operation include modifying clock speeds of the processor and modifying operating voltages of the processor. Reporting of a measured temperature may be utilized, for among other things, to provide feedback to the processor.
For processors utilizing software control mechanisms triggered by management systems such as Advanced Configuration and Power Interface (“ACPI”; Revision 3.0, September 2004; Intel Corp., et al.) or System Management Mode (“SMM”), direct on-die temperature reporting is particularly advantageous, since the ability to report temperatures through the processor to native software improves response time, and provides system designers and manufacturer increased flexibility for adapting processor operation. Flexibility can be further increased by using software-programmable temperature-responsive triggers, which may, for example, be programmed by software and/or firmware instructions executed on the processor (e.g., by third-party software, the operating system, and/or the Basic Input/Output System (“BIOS”)) or be programmed by fuse or other firmware on the die (e.g., programmed by a manufacturer or distributor).
Prior to fabrication, at least one hotspot of the processor is identified by either modeling, empirical data, analysis of the package, and/or by other conventional techniques. A temperature sensor is fabricated in the hotspot, modifying the layout of the circuit to provide for the sensor, if necessary. The sensor may be any temperature sensitive device, such as a p-n junction device (e.g., a diode or transistor) or a temperature sensitive resistor. Advantages of P-N junction devices, besides being easy to integrate onto the processor, are a very quick response to temperature changes and that P-N junction devices do not need a reference temperature (i.e., in comparison to thermocouples). In the embodiments herein, the sensors shown are diodes. However, it should be understood that diodes are only used as an example and that other sensor-types may be also be used.
A first embodiment of the invention is illustrated in
The voltage across diode 130 is provided to a thermal trigger circuit 140 and a measurement circuit 160. In the thermal trigger circuit 140, the diode voltage is compared with a critical point voltage 142 by analog comparator 144. The critical point voltage 142 corresponds to a threshold temperature, such as a maximum thermal operating limit for the processor. By utilizing a simple circuit such as an analog comparator to determine temperature, threshold reporting is near-instantaneous.
Critical point 142 can be fixed or adjustable. A way to provide critical point 142 is by using a digital-to-analog converter to read a register holding a temperature value (not illustrated). The output of the digital-to-analog converter is then provided as input to comparator 144. By programming the register, the threshold can be adjusted.
The thermal trigger signal, output by comparator 144, is provided to thermal control circuits 112 and/or 113. The thermal trigger signal may be provided as an output 101 from the die 100. When the temperature signal (i.e., voltage) from the diode 130 equals or exceeds the critical point 142, the thermal control circuits 112/113 initiate on-die thermal protection, such as a reduction of operating voltage or a reduction of operating frequency. Additionally, the thermal trigger signal may be output from the die to control external thermal control apparatus.
Measurement circuit 150 also receives the temperature signal from the diode 130. An analog-to-digital converter 160 of measurement circuit 150 converts the diode voltage into an M-bit digital value. This M-bit digital value serves as a raw measured temperature reading. The processor 110 may include a software readable register 111, to receive the measured temperature reading. An example of the software readable register 111 is model-specific register (MSR) address space. Alternatively, a code converter 195 can be included to include the raw measured temperature reading into a formatted temperature, which is them provided to the software readable register 111. Use of the code converter 195 allows for compensation of any non-linearity in the raw value, and allows the measured temperature value to be scaled (e.g., Celsius, Fahrenheit). A simple implementation for code converter 195 is a non-volatile memory look-up table.
The measured temperature reading from measurement circuit 150 is also provided to a thermal event detection circuit 180, which directs the processor to modify processor-operation based upon the measured temperature reading. As shown in
As can be understood from the time delays inherent in the circuitry involved, even if the critical point 142 and a trip point (683/685/687/689) are set to a same threshold value, the thermal trigger 140 will respond more quickly than event circuit 180. Accordingly, the critical point 140 may be set for a maximum temperature threshold value the processor can tolerate, since the speed of response can impact processor survivability.
Thermal event detection circuit 180 also may include a circuit 690 directing the processor according to trends of the measured temperature reading over time, and a circuit 692 directing the processor based on an average temperature over time. An example of circuit 690 is at least one memory element to register temperature at a previous time or times and a comparison circuit to compare this previous temperature or temperatures to the current temperature. An example of circuit 392 is at least one memory element to accumulate an average result and at least one computation element to modify the average result in accordance to current temperature, such as a differentiator sampling the value every period N and calculating Trand=T(n)−T(n−1).
Measurement circuit 250 utilizes digital feedback to determine the temperature of the hotspot. Number generator 256 generates a number corresponding to an interim temperature value within a range of temperatures. The interim temperature value is converted into an analog voltage by digital-to-analog converter 252, and comparator 254 compares the analog value with the diode voltage. If the hotspot temperature is approximately equal to the interim temperature value, the interim temperature value is stored in latching register 258, and the number generator 256 can either be reset, or number generation can continue, depending upon the particular technique used. Otherwise, another interim temperature is generated.
This can be better understood by considering particular implementations for number generation, such as linear searches and successive approximation (e.g., a, binary search). The simplest example is a linear search using an up-down counter.
For example, consider a counter sweeping the temperature range, with the hotspot temperature being fixed at a temperature within the range. If the hotspot temperature lies between two interim temperatures produced by the number generator 256, the results are not exact. On an up-sweep of the temperature range, the comparator output 254 will undergo a transition when the hotspot temperature equals or exceeds the interim temperature values. Thus, the register 258 will latch a temperature value that is slightly higher than the actual temperature. Additionally, if a down-sweep is performed for the same diode temperature, the result is different than for the up-sweep (off by one). The accuracy is therefore directly dependent upon the interim temperature-granularity provided by number generator, such that the measured temperature is approximate.
Using the output of comparator 254 to latch register 258 is appropriate with search techniques which will result in comparator transitions at approximately the hotspot temperature. However, if a non-linear search is used, such as a binary search, latching the register in this manner will result in errors, since a binary search can under-shoot and over-shoot the actual temperature during the search. In this case, relying on the output of comparator 254 as the trigger of latching register 258 results in inaccurate readings. Instead, as shown in
Any number of circuits can be used to perform a binary search. For example, a binary search can be performed utilizing two registers, a subtractor, and some basic logic. One register stores a high temperature value, whereas the other register stores a low temperature value. An intermediate value between high and low is determined by subtracting the low temperature value from high temperature value in the subtractor and dividing by 2 (e.g., dropping the lowest order bit subtraction result). This intermediate value is output as the interim temperature value. If the comparator 254 indicates that the hotspot is hotter than the interim temperature value, then low temperature value is set to the intermediate temperature and the process is repeated. Likewise, if the comparator 254 indicates that the hotspot is cooler than the interim temperature value, then high temperature value is set to the intermediate temperature and the process is repeated. The process ends with a triggering of the latching register when the low register and high register lock (i.e., are equal or approximately equal), such that the search is over. Other types of searches can be performed in a similar manner, and lookup table or an arithmetic logic unit (“ALU”) can also be used to determine the intermediate value, instead of the binary subtractor.
Measurement circuits 250 and 350 may also signal the processor and/or thermal control circuitry if the temperature sensed by diode 130 is outside the range of temperatures searched. Examples of how this can be accomplished include: independently comparing the temperature to the high and low values of the range; for a linear search, providing logic utilizing an overflow flag signal of the number generator (e.g., counter) in combination with the comparator 254 output if the full range is swept but no result found; and for a non-linear search, providing logic to signal if an intermediate search temperature value reaches the high or low value, in combination with the comparator 254 output. In such a case, the measured temperature reading is set to be a limit of the range of temperatures closest to the temperature sensed by the diode.
A disadvantage of generating the thermal trigger using events circuit 380 or thermal trigger 240, instead of thermal trigger circuit 140, is a slight delay in response time. An advantage, however, is a reduction analog circuits, which are more sensitive to temperature and transients than digital circuits, usually operate at higher voltages, and generally require more die space. Other advantages are the relative ease with which digital logic can be programmed by software (e.g., without requiring a digital-to-analog converter (“DAC”) to program critical point 142 in thermal trigger circuit 140), and that software-programmable logic can be used instead of hard-wired circuits (thereby increasing die design flexibility).
As an example of operation, a thermal event can initiate a change of the processor frequency if the temperature is high. A policy of the central management system 710 could be to reduce both core's frequency if either core is high. Another, more complex, policy would be to first reduce one core frequency only. If after some time temperature does not go down, the other core frequency and operating voltage would also be reduced. Implementation of this second policy can be accomplished by having a timer feedback to the routing matrix. When the decision table indicates to reduce operations in one processor but not the other, the timer begins counting. When the timer expires, a signal is fed back to the input of the routing matrix, resulting in a reduced operation for the second processor. The delay time can either be fixed, pre-stored in a register, or output by the routing matrix.
Logic circuit 715 compares the measured temperature readings from the plurality of measurement circuits, and provides central management system 710 with a highest of the measured temperature readings as a maximum measured temperature value of the die.
In addition, the central management system 710 can provide central reporting capability. Each core normally can access only its own register set. Central management system 710 can report to each core the package state such as the maximum measured temperature, status bits, etc. Additionally, lookup table entries within central management system 710 can provide some or all of the values used as programmed thresholds for both critical point and thermal events, adapting system thresholds during operation. Central management system 710 can also provide control signals at output 1104 for off-die thermal management.
While the above embodiments utilize a diode for sensing a temperature within a hotspot, any p-n junction device that is sensitive to temperature can be used.
In the embodiments, interrupts used to direct the processor or processors include hardware interrupts, software interrupts such as operating system interrupts, and combinations thereof. Among other things, processor interrupts direct to the operating system may be used to generate native instructions.
The p-n junctions are biased using a band-gap based device to maintain consistent operation as package temperatures change. An example of a band-gap current source for biasing the diodes is a band-gap comparator.
As can be understood from the above description, components of the various embodiments can be interchanged as a matter of design. It is contemplated that numerous modifications may be made to the present invention without departing from the spirit and scope of the invention as defined in the following claims.
1. An apparatus comprising:
- a plurality of processor cores on a semiconductor die, the processor cores having at least one corresponding temperature sensor;
- circuitry on the semiconductor die to generate thermal event indications based on sensed temperatures from multiple temperature sensors of multiple processor cores, a thermal event indication to indicate that a sensed temperature exceeds a temperature point; and
- central management logic on the semiconductor die to receive thermal event indications based on sensed temperatures from multiple temperature sensors of multiple processor cores,
- the central management logic to modify operation of one or more of the processor cores in response to a thermal event indication.
2. The apparatus of claim 1, wherein a processor core has a temperature sensor to sense temperature at a hotspot.
3. The apparatus of claim 1, wherein temperature sensors have one or more corresponding temperature points, and
- wherein the circuitry to generate thermal event indications is to compare sensed temperatures from multiple temperature sensors to corresponding temperature points.
4. The apparatus of claim 1, wherein the circuitry to generate thermal event indications is to compare sensed temperatures from multiple temperature sensors of multiple processor cores to the same temperature point.
5. The apparatus of claim 1, comprising a register on the semiconductor die to store a digital value corresponding to a sensed temperature.
6. The apparatus of claim 1, wherein the circuitry to generate thermal event indications is to compare a digital value corresponding to a sensed temperature to a digital value corresponding to a temperature point.
7. The apparatus of claim 1, wherein the circuitry to generate thermal event indications is to compare sensed temperatures from multiple temperature sensors of multiple processor cores to programmable temperature points.
8. The apparatus of claim 1, the central management logic to modify operation of a processor core from which a sensed temperature generates a thermal event indication.
9. The apparatus of claim 1, the central management logic to generate an interrupt to one or more of the processor cores in response to a thermal event indication.
10. The apparatus of claim 1, the central management logic to modify a clock for one or more of the processor cores in response to a thermal event indication.
11. A method comprising:
- sensing temperatures with multiple temperature sensors of multiple processor cores on a semiconductor die, wherein the processor cores have at least one corresponding temperature sensor;
- generating by circuitry on the semiconductor die thermal event indications based on sensed temperatures from multiple temperature sensors of multiple processor cores, wherein a thermal event indication is to indicate that a sensed temperature exceeds a temperature point;
- receiving by central management logic on the semiconductor die thermal event indications based on sensed temperatures from multiple temperature sensors of multiple processor cores; and
- modifying by the central management logic operation of one or more of the processor cores in response to a thermal event indication.
12. The method of claim 11, wherein sensing temperatures includes sensing a temperature at a hotspot.
13. The method of claim 11, wherein temperature sensors have one or more corresponding temperature points, and
- wherein generating thermal event indications includes comparing sensed temperatures from multiple temperature sensors to corresponding temperature points.
14. The method of claim 11, wherein generating thermal event indications includes comparing sensed temperatures from multiple temperature sensors of multiple processor cores to the same temperature point.
15. The method of claim 11, comprising storing in a register on the semiconductor die a digital value corresponding to a sensed temperature.
16. The method of claim 11, wherein generating thermal event indications includes comparing a digital value corresponding to a sensed temperature to a digital value corresponding to a temperature point.
17. The method of claim 11, wherein generating thermal event indications includes comparing sensed temperatures from multiple temperature sensors of multiple processor cores to programmable temperature points.
18. The method of claim 11, wherein modifying operation includes modifying operation of a processor core from which a sensed temperature generates a thermal event indication.
19. The method of claim 11, wherein modifying operation includes generating an interrupt to one or more of the processor cores in response to a thermal event indication.
20. The method of claim 11, wherein modifying operation includes modifying a clock for one or more of the processor cores in response to a thermal event indication.
International Classification: G06F 1/00 (20060101);