TWI569136B - Temperature estimating method of microprocessor and computer system using the same - Google Patents
Temperature estimating method of microprocessor and computer system using the same Download PDFInfo
- Publication number
- TWI569136B TWI569136B TW103125528A TW103125528A TWI569136B TW I569136 B TWI569136 B TW I569136B TW 103125528 A TW103125528 A TW 103125528A TW 103125528 A TW103125528 A TW 103125528A TW I569136 B TWI569136 B TW I569136B
- Authority
- TW
- Taiwan
- Prior art keywords
- temperature
- functional blocks
- microprocessor
- power consumption
- processing unit
- Prior art date
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Semiconductor Integrated Circuits (AREA)
Description
本發明是有關於一種方法及其電腦系統，且特別是有關於一種電腦系統中微處理器的溫度推算方法。 The present invention relates to a method and computer system thereof, and more particularly to a temperature estimation method for a microprocessor in a computer system.
隨著近代技術進步，目前在微處理器上所容納的電晶體密度迅速上升，伴隨而來的現象為微處理器上單位面積消耗功率大幅上升，不僅導致微處理器平均運作溫度上升，並且還導致微處理器運作時溫度梯度加劇以及諸多區域熱點(Hot Spot)出現，這些現象最後會引發微處理器的漏電流(Leakage Current)上升、時脈偏移(Clock Skew)、系統故障(System Failure)、可靠度(Reliability)及性能(Performance)降低。因而近幾年諸多動態溫度管理(Dynamic Thermal Management,DTM)技術被提出以解決上述所提及問題，然則這些技術皆需取得精確的溫度分佈資訊做為決策基礎。 With the advancement of modern technology, the density of transistors currently contained in microprocessors has risen rapidly. The accompanying phenomenon is that the power consumption per unit area of the microprocessor has risen sharply, which not only causes the average operating temperature of the microprocessor to rise, but also This leads to an increase in temperature gradients during the operation of the microprocessor and the appearance of hot spots in the area. These phenomena can eventually lead to leakage current (Leakage Current), clock skew (Clock Skew), system failure (System Failure). ), reliability (Reliability) and performance (Performance) are reduced. Therefore, in recent years, many Dynamic Temperature Management (DTM) technologies have been proposed to solve the above mentioned problems, but these technologies need to obtain accurate temperature distribution information as the basis for decision making.
微處理器上的溫度管控技術可有效提升其可靠度與效能 及降低漏電效應等，這些技術皆須取得精確及精準的溫度訊息以做正確決策。目前多採用由額外設置於微處理器上的溫度感測器取得溫度數值，但有諸多限制以至於會使溫度管控無法做出正確決策。另一方面溫度訊息可由溫度模型運算而得，然而目前模型皆過於複雜以至於多用於模擬，無法在動態提供訊息。 Temperature control technology on the microprocessor can effectively improve its reliability and efficiency And to reduce leakage effects, these technologies must obtain accurate and accurate temperature information to make the right decision. At present, temperature sensors are additionally used by a temperature sensor additionally provided on the microprocessor, but there are many restrictions that make temperature control unable to make a correct decision. On the other hand, the temperature information can be calculated by the temperature model. However, the current models are too complicated to be used for simulation and cannot provide dynamic information.
本發明提供一種微處理器溫度推算方法及其電腦系統，可快速的取得微處理器中各功能區塊的溫度資訊。 The invention provides a microprocessor temperature estimation method and a computer system thereof, which can quickly obtain temperature information of each functional block in a microprocessor.
本發明提出一種電腦系統，其包括微處理器以及處理單元。微處理器包括多個功能區塊以及效能計數器，其中效能計數器耦接於各功能區塊。處理單元耦接於效能計數器。處理單元取得微處理器所對應的溫度模型，其中溫度模型等效於多個低通濾波器結構。處理單元從效能計數器取得第t個取樣時間區間內對應於各功能區塊的效能計數值。處理單元將對應於各功能區塊的效能計數值轉換為消耗功率值。處理單元又根據消耗功率值以及溫度模型計算得到溫度分佈，其中溫度分佈包括對應於第t個取樣時間區間各功能區塊的溫度值。 The invention provides a computer system comprising a microprocessor and a processing unit. The microprocessor includes a plurality of functional blocks and a performance counter, wherein the performance counter is coupled to each functional block. The processing unit is coupled to the performance counter. The processing unit obtains a temperature model corresponding to the microprocessor, wherein the temperature model is equivalent to a plurality of low pass filter structures. The processing unit obtains the performance counter value corresponding to each functional block in the tth sampling time interval from the performance counter. The processing unit converts the performance count value corresponding to each functional block into a power consumption value. The processing unit further calculates a temperature distribution according to the power consumption value and the temperature model, wherein the temperature distribution includes temperature values corresponding to the functional blocks of the tth sampling time interval.
本發明另提出一種微處理器的溫度推算方法，其中微處理器包括多個功能區塊，包括以下步驟。首先，取得微處理器所對應的溫度模型，其中溫度模型等效於多個低通濾波器結構。然後，取得第t個取樣時間區間內對應於各功能區塊的效能計數值。 接著，將對應於各功能區塊的效能計數值轉換為消耗功率值。之後，根據消耗功率值以及溫度模型計算得到溫度分佈，其中溫度分佈包括對應於第t個取樣時間區間各功能區塊的溫度值。 The invention further provides a temperature estimation method for a microprocessor, wherein the microprocessor comprises a plurality of functional blocks, including the following steps. First, the temperature model corresponding to the microprocessor is obtained, wherein the temperature model is equivalent to a plurality of low-pass filter structures. Then, the performance count value corresponding to each functional block in the tth sampling time interval is obtained. Next, the performance count value corresponding to each functional block is converted into a power consumption value. Thereafter, the temperature distribution is calculated according to the power consumption value and the temperature model, wherein the temperature distribution includes temperature values corresponding to the functional blocks of the tth sampling time interval.
基於上述，本發明所提供的微處理器溫度的推算方法及其電腦系統，使得微處理器不需外加任何溫度或電流的感測器即可快速的推測各功能區塊的溫度資訊。 Based on the above, the method for estimating the temperature of the microprocessor provided by the present invention and the computer system thereof enable the microprocessor to quickly estimate the temperature information of each functional block without adding any temperature or current sensor.
為讓本發明的上述特徵和優點能更明顯易懂，下文特舉實施例，並配合所附圖式作詳細說明如下。 The above described features and advantages of the invention will be apparent from the following description.
10‧‧‧微處理器 10‧‧‧Microprocessor
110‧‧‧晶片 110‧‧‧ wafer
120‧‧‧熱傳導器 120‧‧‧Heat Conductor
130‧‧‧散熱器 130‧‧‧heatsink
140‧‧‧處理單元 140‧‧‧Processing unit
150‧‧‧效能計數器 150‧‧‧performance counter
100‧‧‧電腦系統 100‧‧‧ computer system
CV‧‧‧效能計數值 CV‧‧‧ performance count
B1~B3‧‧‧功能區塊 B1~B3‧‧‧ functional block
R1~R3、Rsp、Rhs、Rco、Rifi、Rifj‧‧‧熱電阻 R1~R3, R sp , R hs , R co , R ifi , R ifj ‧‧‧ Thermal resistance
Rij‧‧‧水平熱電阻 R ij ‧‧‧ horizontal thermal resistance
Ci、Csp、Chs‧‧‧熱電容 C i, C sp, C hs ‧‧‧ thermal capacity
ISi、ISj‧‧‧電流源 IS i , IS j ‧‧‧current source
Ii、Ij‧‧‧電流/輸入功率 I i , I j ‧‧‧current / input power
S201~S204、S601~S608‧‧‧溫度推算方法的流程步驟 S201~S204, S601~S608‧‧‧ Process steps of temperature estimation method
Va‧‧‧周遭溫度 V a ‧ ‧ ambient temperature
圖1為一種習知的微處理器的溫度模型的結構示意圖。 FIG. 1 is a schematic structural diagram of a temperature model of a conventional microprocessor.
圖2為根據本發明一實施例所繪示微處理器的溫度推算方法之流程圖。 FIG. 2 is a flow chart showing a method for estimating a temperature of a microprocessor according to an embodiment of the invention.
圖3為根據本發明一實施例所繪示微處理器的功能方塊圖。 FIG. 3 is a functional block diagram of a microprocessor according to an embodiment of the invention.
圖4為根據本發明一實施例所繪示兩個相鄰的功能區塊之溫度模型的等效電路圖。 4 is an equivalent circuit diagram showing temperature models of two adjacent functional blocks according to an embodiment of the invention.
圖5為根據本發明一實施例所繪示第i個功能區塊之溫度模型的的等效電路圖。 FIG. 5 is an equivalent circuit diagram showing a temperature model of an i-th functional block according to an embodiment of the invention.
圖6為根據本發明一實施例所繪示微處理器的溫度推算方法的流程圖。 FIG. 6 is a flow chart showing a method for estimating a temperature of a microprocessor according to an embodiment of the invention.
一般而言，評估微處理器運作時的溫度的分佈及變化之方法主要可分為感測器偵測法(Sensor-based Approaches)以及推算法(Emulation-based Approaches)。 In general, methods for evaluating the distribution and variation of temperature during operation of a microprocessor are mainly classified into Sensor-based Approaches and Emulation-based Approaches.
使用感測器偵測法時，微處理器上需額外設置一或多個溫度感測器(Thermal Sensor)或是電流感測器(Current Sensor)，其後再於微處理器運作期間讀取這些溫度感測器或電流感測器的量測值。然而，溫度感測器或電流感測器偵測法具有諸多限制，例如，就現有技術而言，溫度感測器以及電流感測器所能達到之量測精準度為±1℃，而這樣±1℃的誤差將會影響微處理器的動態溫度管理之策略效率。再者，於微處理器上嵌入溫度感測器或電流感測器則會導致微處理器的晶片面積增大，同時亦使製造成本增加。此外，由於微處理器上的功能方塊(functional blocks)的設置並無法隨意的更動，溫度感測器以及電流感測器亦沒有辦法隨意的設置於微處理器上的任意位置。 When using the sensor detection method, one or more Thermal Sensors or Current Sensors are required on the microprocessor, and then read during the operation of the microprocessor. The measured values of these temperature sensors or current sensors. However, temperature sensor or current sensor detection has many limitations. For example, in the prior art, the temperature sensor and the current sensor can achieve a measurement accuracy of ±1 ° C, and thus An error of ±1 °C will affect the strategic efficiency of the microprocessor's dynamic temperature management. Furthermore, embedding a temperature sensor or current sensor on the microprocessor results in an increase in the wafer area of the microprocessor while also increasing manufacturing costs. In addition, since the setting of the functional blocks on the microprocessor cannot be arbitrarily changed, the temperature sensor and the current sensor are not freely arranged at any position on the microprocessor.
另一方面，使用推算法時，則由微處理器耗能資訊代入各式溫度模型予以進行模擬或運算。就現有技術而言，推算法亦有許多限制。例如，習知模型皆過於複雜進而導致取得溫度資訊時所需計算時間過長且需大量記憶體，難以被即時的用於動態溫度管理。 On the other hand, when using the push algorithm, the microprocessor energy consumption information is substituted into various temperature models for simulation or calculation. As far as the prior art is concerned, the push algorithm has many limitations. For example, the conventional models are too complicated to cause the calculation time to be long when the temperature information is obtained and a large amount of memory is required, which is difficult to be used for dynamic temperature management in real time.
圖1為一種習知的微處理器的溫度模型的結構示意圖。如圖1所示，微處理器10包括了晶片110、熱傳導器(heat spreader) 120以及散熱器(heat sink)130，並且晶片110、熱傳導器120以及散熱器130以圖1所示的方式相疊設置，即，晶片110設置於熱傳導器120上，而熱傳導器120設置於散熱器130之上。而在圖1所示微處理器10中，晶片110上包括了3個功能區塊B1、B2以及B3。在這樣的溫度模型中，微處理器10中的所有熱傳導情形皆被轉化以熱電阻和熱電容來表示熱傳導率的倒數以及熱容。微處理器10的周遭溫度(ambient temperature)則由接地符號表示。例如，在圖1中，熱電阻R1用以表示從功能區塊B1的中心功率輸入點至功能區塊B3的熱傳導率的倒數，而熱電阻R2則用以表示從功能區塊B1的中心功率輸入點至功能區塊B2的熱傳導率的倒數。而熱電阻R3、R4則分別用以表示功能區塊B3的中心功率輸入點至功能區塊B1、B2的熱傳導率的倒數。而熱電阻Rsp、Rhs、以及Rco則分別代表熱傳導器120的熱傳導率的倒數、散熱器130的熱傳導率的倒數，以及散熱器130至周遭環境(對應於標示為接地符號的周遭溫度)的熱傳導率的倒數。 FIG. 1 is a schematic structural diagram of a temperature model of a conventional microprocessor. As shown in FIG. 1, the microprocessor 10 includes a wafer 110, a heat spreader 120, and a heat sink 130, and the wafer 110, the heat conductor 120, and the heat sink 130 are in the manner shown in FIG. The stack is disposed, that is, the wafer 110 is disposed on the heat conductor 120, and the heat conductor 120 is disposed on the heat sink 130. In the microprocessor 10 of FIG. 1, the wafer 110 includes three functional blocks B1, B2, and B3. In such a temperature model, all thermal conduction conditions in the microprocessor 10 are converted to thermal resistance and thermal capacitance to represent the reciprocal of thermal conductivity and heat capacity. The ambient temperature of the microprocessor 10 is represented by a ground symbol. For example, in FIG. 1, the thermal resistor R1 is used to represent the reciprocal of the thermal conductivity from the central power input point of the functional block B1 to the functional block B3, and the thermal resistor R2 is used to represent the central power from the functional block B1. Enter the reciprocal of the thermal conductivity of the point to function block B2. The thermal resistors R3 and R4 are respectively used to represent the reciprocal of the thermal conductivity of the central power input point of the functional block B3 to the functional blocks B1 and B2. The thermal resistances R sp , R hs , and R co represent the reciprocal of the thermal conductivity of the heat conductor 120, the reciprocal of the thermal conductivity of the heat sink 130, and the surrounding environment of the heat sink 130 (corresponding to the ambient temperature indicated as the ground symbol). The reciprocal of the thermal conductivity.
而由圖1所示的微處理器10的溫度模型可知，熱電阻以及熱電容的關係十分複雜，若是以這樣溫度模型對各功能區塊B1~B3的溫度進行推算，則將會耗費相當大的運算資源以及記憶體，因此無法直接使用於動態期間推算得溫度資訊。 The temperature model of the microprocessor 10 shown in FIG. 1 shows that the relationship between the thermal resistance and the thermal capacitance is very complicated. If the temperature of each functional block B1 to B3 is estimated by such a temperature model, it will be quite expensive. The computing resources and memory, so can not be directly used to calculate the temperature information during the dynamic period.
本發明中所提出的微處理器的溫度推算方法由上述藉由微處理器的溫度模型的推算法之概念出發，可快速取得溫度資訊和耗用極小記憶體的特性，使其適用於微處理器運作時，迅速計 算出微處理器的溫度分佈狀況。以下則將透過實施例以及圖示詳細說明。 The temperature estimation method of the microprocessor proposed in the present invention is based on the above-mentioned concept of the temperature model of the microprocessor, and can quickly obtain temperature information and consumes extremely small memory characteristics, making it suitable for micro processing. Quickly calculate when the device is operating Calculate the temperature distribution of the microprocessor. Hereinafter, the embodiment and the drawings will be described in detail.
圖2為根據本發明一實施例所繪示微處理器的溫度推算方法之流程圖。其中，微處理器中包括多個功能區塊(如圖1所示功能區塊B1~B3)。請參照圖2，首先在步驟S201時，取得微處理器材質與環境參數並藉以建立對應的溫度模型，其中溫度模型等效於多個低通濾波器結構。然後在步驟S202時，取得第t個取樣時間區間內對應於各功能區塊的效能計數值。接著在步驟S203時，將對應於各功能區塊的效能計數值轉換為消耗功率值。在步驟S204時，根據消耗功率值以及溫度模型計算得到溫度分佈，其中溫度分佈包括對應於第t個取樣時間區間各功能區塊的溫度值。 FIG. 2 is a flow chart showing a method for estimating a temperature of a microprocessor according to an embodiment of the invention. The microprocessor includes a plurality of functional blocks (such as the functional blocks B1 to B3 shown in FIG. 1). Referring to FIG. 2, first, in step S201, the microprocessor material and the environmental parameters are obtained and a corresponding temperature model is established, wherein the temperature model is equivalent to a plurality of low-pass filter structures. Then, in step S202, the performance count value corresponding to each functional block in the tth sampling time interval is obtained. Next, at step S203, the performance count value corresponding to each functional block is converted into a power consumption value. At step S204, a temperature distribution is calculated according to the power consumption value and the temperature model, wherein the temperature distribution includes temperature values corresponding to the functional blocks of the tth sampling time interval.
圖3為根據本發明一實施例所繪示電腦系統的功能方塊圖。請參照圖3，在本實施例中，電腦系統100包括微處理器10以及處理單元140。微處理器10則包括了效能計數器150，以及多個功能區塊(如圖1所示功能區塊B1~B3)。效能計數器150耦接各功能區塊，而電腦系統100的處理單元140耦接效能計數器150。 FIG. 3 is a functional block diagram of a computer system according to an embodiment of the invention. Referring to FIG. 3, in the embodiment, the computer system 100 includes a microprocessor 10 and a processing unit 140. The microprocessor 10 includes a performance counter 150 and a plurality of functional blocks (such as the functional blocks B1 B B3 shown in FIG. 1). The performance counter 150 is coupled to each functional block, and the processing unit 140 of the computer system 100 is coupled to the performance counter 150.
處理單元140取得微處理器10所對應的溫度模型，其中溫度模型等效於多個低通濾波器結構。處理單元從效能計數器取得第t個取樣時間區間內對應於各功能區塊的多個效能計數值CV。處理單元將對應於各功能區塊的效能計數值CV轉換為多個 消耗功率值。以及，處理單元根據消耗功率值以及溫度模型計算得到溫度分佈，其中溫度分佈包括對應於第t個取樣時間區間各功能區塊的溫度值。 The processing unit 140 obtains a temperature model corresponding to the microprocessor 10, wherein the temperature model is equivalent to a plurality of low pass filter structures. The processing unit obtains, from the performance counter, a plurality of performance count values CV corresponding to the respective functional blocks in the t-th sampling time interval. The processing unit converts the performance count value CV corresponding to each functional block into multiple Power consumption value. And, the processing unit calculates a temperature distribution according to the power consumption value and the temperature model, wherein the temperature distribution includes temperature values corresponding to the functional blocks of the tth sampling time interval.
在本實施例中，處理單元140可為圖1所示微處理器10中晶片110的主要處理單元(並對應於區塊B1~B3之一)，而處理單元140於運作上述溫度推算方法外亦可同時/分時的方式執行其他工作(task)。處理單元140亦可為獨立建置於晶片110中的處理單元，僅針對上述溫度推算方法進行運算及處理，並將得到的結果(例如微處理器10的溫度分佈)傳送至晶片110中的主要處理單元。 In this embodiment, the processing unit 140 may be the main processing unit of the wafer 110 in the microprocessor 10 shown in FIG. 1 (and corresponding to one of the blocks B1 B B3), and the processing unit 140 operates outside the temperature estimation method. Other tasks can also be performed in a simultaneous/time-sharing manner. The processing unit 140 can also be a processing unit independently built in the wafer 110, and only performs operations and processing on the above temperature estimation method, and transmits the obtained result (for example, the temperature distribution of the microprocessor 10) to the main in the wafer 110. Processing unit.
在本發明中，微處理器所對應的溫度模型即包括了各功能區塊的於熱傳導情形時的等效電路。而在本實施例中，各功能區塊的等效電路經過簡化而成為等同於多階低通濾波器之結構。以下則將配合圖式說明此簡化過程。 In the present invention, the temperature model corresponding to the microprocessor includes the equivalent circuit of each functional block in the case of heat conduction. In the present embodiment, the equivalent circuit of each functional block is simplified to become equivalent to the structure of the multi-stage low-pass filter. The simplification process will be described below in conjunction with the schema.
圖4為根據本發明一實施例所繪示兩個相鄰的功能區塊之溫度模型的等效電路圖。此兩相鄰的功能區塊在此實施例中以第i個區塊以及第j個區塊作為表示，可對應於圖1所示結構中功能區塊B1~B3中之兩個功能區塊。 4 is an equivalent circuit diagram showing temperature models of two adjacent functional blocks according to an embodiment of the invention. In this embodiment, the two adjacent functional blocks are represented by the i-th block and the j-th block, and may correspond to two functional blocks in the functional blocks B1 B B3 in the structure shown in FIG. 1 . .
請參照圖4，在第i個功能區塊中，具有提供電流Ii的電流源ISi(電流Ii即為輸入功率Ii)、其內部的熱電阻Rifi(即對應於第i個功能區塊的介面材料的熱傳導率的倒數)，以及與周遭環境之間的熱電容Ci(對應於第i個功能區塊與周遭環境間的熱 容)。在第j個功能區塊中，具有提供電流Ij的電流源ISj(電流Ij即為輸入功率Ij)、其內部的熱電阻Rifj(即對應於第j個功能區塊的介面材料的熱傳導率的倒數)，以及與周遭環境之間的熱電容Cj(對應於第j個功能區塊與周遭環境間的熱容)。而第i個功能區塊以及第j個功能區塊間則包括了電阻Rij，用來表示第i個功能區塊以及第j個功能區塊之間的熱傳導率之倒數。第i個功能區塊以及第j個功能區塊並同時透過熱電阻Rsp、Rhs以及Rco連接至周遭環境。熱電阻Rsp、Rhs以及Rco則分別與圖1所示相同地，為熱傳導器120的熱傳導率的倒數、散熱器130的熱傳導率的倒數，以及散熱器130至周遭環境(對應於標示為接地符號的周遭溫度Va)的熱傳導率的倒數。熱傳導器120以及散熱器130則分別亦具有與周遭環境之間的熱電容Csp、Chs。 Referring to FIG. 4, in the i-th functional block, the current source IS i (current I i is the input power I i ) and the internal thermal resistance R ifi (ie corresponding to the i-th function) are provided. The reciprocal of the thermal conductivity of the interface material of the block, and the thermal capacitance C i (corresponding to the heat capacity between the i-th functional block and the surrounding environment) with the surrounding environment. In the jth functional block, there is a current source IS j that supplies current I j (current I j is input power I j ), and its internal thermal resistance R ifj (ie, interface corresponding to the jth functional block) The reciprocal of the thermal conductivity of the material, and the thermal capacitance C j (corresponding to the heat capacity between the jth functional block and the surrounding environment) with the surrounding environment. The ith functional block and the jth functional block include a resistor R ij for indicating the reciprocal of the thermal conductivity between the i-th functional block and the j-th functional block. The i-th functional block and the j-th functional block are simultaneously connected to the surrounding environment through the thermal resistors R sp , R hs and R co . The thermal resistances R sp , R hs , and R co are the reciprocal of the thermal conductivity of the heat conductor 120, the reciprocal of the thermal conductivity of the heat sink 130, and the ambient temperature of the heat sink 130 (corresponding to the indication), respectively, as shown in FIG. The reciprocal of the thermal conductivity of the ambient temperature V a ) of the ground symbol. The heat conductor 120 and the heat sink 130 also have thermal capacitances C sp and C hs respectively with the surrounding environment.
在此溫度模型中，熱傳導的流向可分為兩類，其一為垂直熱傳導路徑，其為由熱源直接傳導至周邊溫度(Ambient Temperature)的路徑，例如由熱電阻Rifi、Rsp、Rhs以及Rco所串接而成的熱傳導路徑。而另一則為水平熱傳導路徑，其為各功能區塊間傳導路徑，例如熱電阻Rij於第i個功能區塊以及第j個功能區塊之間的的熱傳導路徑。由於微處理器中的矽晶片層(對應於圖1所示晶片110)厚度極薄，這樣將使得垂直熱傳導的效率會遠高於水平熱傳導。而由於垂直熱傳導的效率會遠高於水平熱傳導。因此於等效電路中(如圖4所示)，水平熱電阻將遠大於垂直熱電阻總和。 In this temperature model, the flow direction of heat conduction can be divided into two types, one of which is a vertical heat conduction path, which is a path that is directly conducted by a heat source to an ambient temperature, such as a thermal resistance R ifi , R sp , and R hs . And a heat conduction path in which R co is connected in series. And the other level was heat conduction path, which is a conductive path between the various functional blocks, such as thermal resistance R ij heat conduction path between the i-th and j-th functional blocks of the functional blocks. Since the germanium wafer layer (corresponding to the wafer 110 shown in Figure 1) in the microprocessor is extremely thin, this will make the vertical heat transfer efficiency much higher than the horizontal heat transfer. And because the efficiency of vertical heat conduction is much higher than the horizontal heat conduction. Therefore, in the equivalent circuit (as shown in Figure 4), the horizontal thermal resistance will be much larger than the sum of the vertical thermal resistances.
因此，在本實施例中，將先忽略第i個功能區塊間以及第j個功能區塊間的水平電阻(即，熱電阻Rij)並予以開路。這樣的運算方式會使得推算的溫度具有一定的誤差，但經實驗證實這樣的誤差仍於可容許的範圍內。接著，在將水平電阻開路後，各個功能區塊(例如，第i個功能區塊間以及第j個功能區塊)則將形成多階低通濾波器(Multi-order Low-pass Filter)結構，例如N階低通濾波器，其中N為大於等於1的正整數，並且因而可使用多階低通濾波器等效公式進行運算溫度狀態。 Therefore, in the present embodiment, the horizontal resistance (i.e., the thermal resistance R ij ) between the i-th functional block and the j-th functional block will be ignored and opened. Such an operation method causes the estimated temperature to have a certain error, but it has been experimentally confirmed that such an error is still within an allowable range. Then, after the horizontal resistance is opened, each functional block (for example, the i-th functional block and the j-th functional block) will form a multi-order low-pass filter structure. For example, an N-th order low-pass filter, where N is a positive integer greater than or equal to 1, and thus the operating temperature state can be performed using a multi-order low-pass filter equivalent formula.
然而由於微處理器中包含管線(Pipeline)及多工(Multi-task)技術，所以每個功能區塊的功率消耗隨時間變化甚劇，使得輸入電流源Ii、Ij可被視為高頻信號，所以在多階低通濾波器主宰高頻訊號的電容(即，熱電容Ci)會有顯著效應，而其餘電容(例如，熱電容Csp以及Chs)影響較小。因此，在本實施例中，將僅考慮熱電容Ci(熱電容Cj)的效率而忽略熱電Csp以及Chs的影響，使得第i個功能區塊(第j個功能區塊)形成為單階的濾波器結構。 However, since the microprocessor includes pipeline (Pipeline) and multi-task technology, the power consumption of each functional block changes with time, so that the input current sources I i , I j can be regarded as high. The frequency signal, so the capacitance of the high-frequency signal (ie, the thermal capacitance C i ) in the multi-stage low-pass filter has a significant effect, while the remaining capacitances (for example, the thermal capacitance C sp and C hs ) have less influence. Therefore, in the present embodiment, only the efficiency of the thermal capacitance C i (thermal capacitance C j ) is considered, and the influence of the thermoelectric C sp and C hs is ignored, so that the i-th functional block (j-th functional block) is formed. It is a single-order filter structure.
經由上述對熱電阻和熱電容的簡化後，第i個功能區塊的等效電路圖即如圖5所示。圖5為根據本發明一實施例所繪示第i個功能區塊之溫度模型的的等效電路圖。請參照圖5，在本實施例中，第i個功能區塊之溫度模型的的等效電路即為一單階濾波器結構，具有並聯的熱電阻Rvi和熱電容Ci，其中熱電阻Rvi即為圖3所示之熱電阻Rifi、Rsp、Rhs及Rco所串聯而成。電流源ISi所提供
的電流Ii即為輸入功率，而電流IRvi及電流ICi分別可視為分流至熱電阻Rvi及熱電容Ci的功率。藉此，輸入功率(即，電流Ii)即可被表示同下式(1)：I i =I Rvi +I Ci (1)轉換上式(1)後，電流IRvi則可被表示為下式(2)：
值得注意的是，在式(4)中所需要的參數包括了取樣時間區間t的輸入功率(即，電流I i (t))以及前一個取樣時間區間(t-△t)分流至熱電阻Rvi的功率(即，電流I Rvi (t-△t))以及熱電阻Rifi、Rsp、Rhs及Rco、熱電容Ci等參數。其中，熱電阻Rifi、Rsp、Rhs及Rco和熱電容Ci和周邊溫度Va的數值則可根據微處理器10中晶片110、熱傳導器120以及散熱器130的材料之材料參數以及厚度，和當時的環境溫度而取得。微處理器10中的處理單元140(如圖3所示)則可預先根據上述的參數(材料參數以及厚度和當時的環境溫度)得出熱電阻、熱電容及周邊溫度數值。 It is worth noting that the parameters required in equation (4) include the input power of the sampling time interval t (ie, current I i ( t )) and the previous sampling time interval ( t −Δ t ) to the thermal resistance. R vi power (ie, current I Rvi ( t - Δ t )) and thermal resistance R ifi , Rsp, R hs and R co , thermal capacitance C i and other parameters. The values of the thermal resistances R ifi , R sp , R hs and R co and the thermal capacitance C i and the ambient temperature V a may be based on the material parameters of the materials of the wafer 110, the heat conductor 120 and the heat sink 130 in the microprocessor 10. And the thickness, and the ambient temperature at that time. The processing unit 140 (shown in FIG. 3) in the microprocessor 10 can derive the values of the thermal resistance, the thermal capacitance, and the ambient temperature based on the above parameters (material parameters and thickness and current ambient temperature).
而取樣時間區間t的輸入功率則可由各功能區塊的多個效能計數值CV轉換而得(對應於圖2步驟S202)，前一個取樣時間區間(t-△t)分流至熱電阻Rvi的功率則取至於前一個取樣時間區間運算結果。各個功能區塊由於功能之不同，而可能具有不同程度的功率消耗能力。在本發明一實施例中，處理單元140中預先建立了各個功能區塊的消耗功率與效能計數值的轉換表，當處理單元從效能計數器150取得效能計數值CV，處理單元140即可透過轉換表以一查表方式將對應於各個功能區塊的效能計數值CV轉換為對應於各個功能區塊的消耗功率。而各個功能區塊的消耗功率即為上述溫度模型中的輸入功率。因此，消耗功率即可直接代入上述的式(3)的電流I i (t)當中，各個功能區塊於取樣時間區 間t的溫度值(例如第i個功能區塊於取樣時間區間t的溫度值Vi(t))即可被求得(對應於圖2所示步驟S203)。各個功能區塊於取樣時間區間t的溫度值即形成微處理器10於取樣時間區間t之溫度分佈。再者，在取樣時間區間t之分流至熱電阻Rvi的功率亦將暫存於處理單元中一記憶體(未繪示)中，待下一個時間點(例如時間點(t+△t))時計算取樣時間區間(t+△t)的溫度值時可被運用。 T is the sampling time interval can be input by the plurality of power performance counter value CV obtained by conversion of the functional blocks (FIG. 2 corresponds to the step S202), before one sampling time interval (t - △ t) to a thermal shunt resistor R vi The power is taken as the result of the previous sampling time interval. Each functional block may have different degrees of power consumption due to different functions. In an embodiment of the present invention, the processing unit 140 pre-establishes a conversion table of the power consumption and the performance count value of each functional block. When the processing unit obtains the performance count value CV from the performance counter 150, the processing unit 140 can perform the conversion. The table converts the performance count value CV corresponding to each functional block into a power consumption corresponding to each functional block in a lookup table manner. The power consumption of each functional block is the input power in the above temperature model. Therefore, the power consumption can be directly substituted into the current I i ( t ) of the above formula (3), and the temperature value of each functional block in the sampling time interval t (for example, the temperature of the i-th functional block in the sampling time interval t) The value Vi(t) can be obtained (corresponding to step S203 shown in Fig. 2). The temperature value of each functional block in the sampling time interval t forms the temperature distribution of the microprocessor 10 in the sampling time interval t. Further, the sampling time interval t is the power split to the thermal resistance R vi processing will also be temporarily stored in a memory unit (not shown), until the next time point (e.g., a time point (t + △ t) can be calculated using the time when the sampling time interval) (t + △ t) temperature value.
另一方面，為了使得微處理器10的溫度分佈更為精確，在本發明一實施例中，在上述圖5所示實施例中被省略的水平熱電阻(例如第i個功能區塊以及第j個功能區塊間的水平熱電阻Rij)則被重新併入計算。 On the other hand, in order to make the temperature distribution of the microprocessor 10 more precise, in one embodiment of the present invention, the horizontal thermal resistance (for example, the i-th functional block and the The horizontal thermal resistance R ij ) between the j functional blocks is reincorporated into the calculation.
簡單來說，將水平熱電阻所造成的熱傳導效應將併入式(3)中則可得到下式(5)：
值得注意的是，於圖5所示的實施例中，為了將各個功能區塊之溫度模型的等效電路簡化為單階低通濾波器結構時，除了省略上述提到的水平熱傳導效應的考量外，亦同時省略了熱電容Csp以及Chs對溫度所產生的效應。因此，在本發明一實施例中，亦可再考慮水平熱傳導效應外，亦將熱電容Csp以及Chs對溫度所產生的效應併入溫度分佈的計算。然而，如圖5所示的等效電路在併入熱電容Csp以及Chs所產生的效應之考量後，將形成多階的低通濾波器結構，其所對應的溫度值計算將比圖5所示的單階低通濾波器結構之等效電路複雜許多。因此，是否併入熱電容Csp以及Chs所產生的效應之考量則可視實際實施時的狀況而定。 It should be noted that in the embodiment shown in FIG. 5, in order to simplify the equivalent circuit of the temperature model of each functional block into a single-stage low-pass filter structure, in addition to omitting the above-mentioned horizontal heat conduction effect considerations In addition, the effects of the thermal capacitances C sp and C hs on the temperature are also omitted. Therefore, in one embodiment of the present invention, in addition to the horizontal heat conduction effect, the effects of the thermal capacitances C sp and C hs on temperature are also incorporated into the calculation of the temperature distribution. However, the equivalent circuit shown in FIG. 5 will form a multi-stage low-pass filter structure after considering the effects of the thermal capacitances C sp and C hs , and the corresponding temperature values will be calculated. The equivalent circuit of the single-stage low-pass filter structure shown in 5 is much more complicated. Therefore, consideration of whether or not the effects of the thermal capacitances C sp and C hs are incorporated may be determined depending on the actual implementation.
圖6為根據本發明一實施例所繪示微處理器的溫度推算方法的流程圖。相較於圖2所示實施例，圖6所示實施例提供了一種較為詳細的實施方式。其中，圖2中步驟S201可對應於圖6 所示實施例中步驟S601~S603，步驟S202~S203可對應於圖6所示實施例中步驟S604，而步驟S204則可對應於圖6所示實施例中步驟S605~S608。 FIG. 6 is a flow chart showing a method for estimating a temperature of a microprocessor according to an embodiment of the invention. Compared to the embodiment shown in FIG. 2, the embodiment shown in FIG. 6 provides a more detailed implementation. Wherein, step S201 in FIG. 2 may correspond to FIG. 6 In the embodiment, steps S601-S603, steps S202-S203 may correspond to step S604 in the embodiment shown in FIG. 6, and step S204 may correspond to steps S605-S608 in the embodiment shown in FIG.
請參照圖6以及圖2，首先，處理單元140將讀取環境參數(步驟S601)，例如微處理器10中各元件之材料參數以及厚度，和當時的環境溫度等，並根據環境參數以及微處理器10中各功能區塊的分佈以及相鄰情形建立微處理器10的溫度模型(步驟S602)。接著，處理單元140則根據上述微處理器10的溫度模型計算各功能區塊對應於式(4)中的α值(步驟S603)。其中，α值可由各功能區塊中的熱電阻以及熱電容所計算得到，並且對應於各功能區塊前一時間點的輸入功率與目前的輸出功率與各功能區塊目前的溫度值所造成的影響之比例關係。 Referring to FIG. 6 and FIG. 2, first, the processing unit 140 will read the environmental parameters (step S601), such as the material parameters and thickness of each component in the microprocessor 10, and the current ambient temperature, etc., and according to the environmental parameters and micro The distribution of the functional blocks in the processor 10 and the adjacent conditions establish a temperature model of the microprocessor 10 (step S602). Next, the processing unit 140 calculates, according to the temperature model of the microprocessor 10 described above, each functional block corresponds to the alpha value in the equation (4) (step S603). Wherein, the value of α can be calculated from the thermal resistance and the thermal capacitance in each functional block, and corresponds to the input power at the previous time point of each functional block and the current output power and the current temperature value of each functional block. The proportional relationship of the impact.
然後，處理單元140則從效能計數器150讀取對應於各功能區塊的效能計數值，並以查表方式根據功能區塊的功率消耗能力轉換各功能區塊的效能計數值為各功能區塊的消耗功率值(步驟S604)。接著，根據式(4)，處理單元140即可取得在目前(即，取樣時間區間t內，或稱第t個取樣時間區間內)內各個功能區塊的溫度值，也就是微處理器10於取樣時間區間t內的溫度分佈(步驟S605)。 Then, the processing unit 140 reads the performance count value corresponding to each functional block from the performance counter 150, and converts the performance count value of each functional block according to the power consumption capability of the functional block in a table lookup manner. The power consumption value (step S604). Then, according to formula (4), the processing unit 140 can obtain the temperature values of the respective functional blocks in the current (ie, within the sampling time interval t, or within the t-th sampling time interval), that is, the microprocessor 10 The temperature distribution in the sampling time interval t (step S605).
再者，處理單元140更可根據水平熱傳導效應更新上述於取樣時間區間t內的溫度分佈，也就是針對各功能區塊分別計算得到式(8)中的後兩項後，加入上述的溫度分佈中(S606)。而 此步驟可根據實際狀況選擇性的實施，例如在需要較精確的溫度分佈結果時，或是當處理單元140的於執行對應於溫度推算方法的程序外之工作量較低時實施，但本發明並不限定於上述。 Furthermore, the processing unit 140 may further update the temperature distribution in the sampling time interval t according to the horizontal heat conduction effect, that is, after calculating the latter two items in the equation (8) for each functional block, adding the above temperature distribution. Medium (S606). and This step may be selectively implemented according to actual conditions, for example, when a more accurate temperature distribution result is required, or when the workload of the processing unit 140 outside the program corresponding to the temperature estimation method is low, but the present invention It is not limited to the above.
處理單元140於取得目前的溫度分佈後，則更進一步的判斷是否可從效能計數器150取得下一個取樣時間區間(即，取樣時間區間(t+△t))時各功能區塊的效能計數值。若否，則代表微處理器10已停止運作，可結束對應於溫度推算方法的程序。若是，處理單元140則儲存目前的溫度分佈，並設定取樣時間區間t等於(t+△t)，重新執行步驟S604~S607，以輸出取樣時間區間(t+△t)的溫度分佈。 Post-processing unit 140 to obtain the current temperature distribution, it is further determined whether a sampling time interval from the acquisition performance counter 150 (i.e., the sampling time interval (t + △ t)) for each functional block count value efficacy when . If not, it means that the microprocessor 10 has stopped operating, and the program corresponding to the temperature estimation method can be ended. If so, the processing unit 140 to store the current temperature distribution, and sets the sampling time interval t is equal to (t + △ t), re-executes step S604 ~ S607, a temperature output sample time interval (t + △ t) distribution.
值得注意的是，處理單元並不需要在每次執行對應於微處理器的溫度推算方法的程序時重新執行步驟S601~S603，可於第一次執行對應於溫度推算方法的程序時執行，並且將執行後的結果(對應於各功能區塊的等效電路以及α值等)儲存於處理單元中的記憶體中。 It is to be noted that the processing unit does not need to re-execute steps S601 to S603 each time the program corresponding to the temperature estimation method of the microprocessor is executed, and can be executed when the program corresponding to the temperature estimation method is executed for the first time, and The result after execution (corresponding to the equivalent circuit of each functional block and the alpha value, etc.) is stored in the memory in the processing unit.
綜上所述，本發明所提出了一種微處理器的溫度估算方法以及包括上述微處理器的電腦系統，使用微處理器中效能計數器所輸出的對應於微處理器中各功能區塊的效能計數值，以及簡化後的溫度模型推算微處理器的溫度分佈。由於效能計數器一般已設置於微處理器中，微處理器則不需在為了取得溫度分佈而額外設置溫度感測器或電流感測器。另一方面，由於簡化後的溫度模型可加速溫度分佈的推算，使得上述的溫度分佈可適用於動態 提供溫度訊息，因此，微處理器可以更即時的針對當下的溫度狀況進行處置。並且，經由實驗證明，於本發明中所提出的溫度模型與現有的溫度模型(例如圖1所示)相比，使用本發明所提出的溫度模型更可將推算溫度的精準度之誤差降低於正負0.8℃以內，並且計算速度至少為現有技術的5倍，所需使用的記憶體大小亦為現有技術的0.45%以下。 In summary, the present invention provides a method for estimating the temperature of a microprocessor and a computer system including the above microprocessor, which uses the performance counter of the microprocessor to output the performance corresponding to each functional block in the microprocessor. The count value, as well as the simplified temperature model, estimate the temperature distribution of the microprocessor. Since the performance counter is typically already set in the microprocessor, the microprocessor does not need to additionally provide a temperature sensor or current sensor for temperature distribution. On the other hand, since the simplified temperature model accelerates the estimation of the temperature distribution, the above temperature distribution can be applied to dynamics. The temperature message is provided so that the microprocessor can more quickly handle the current temperature conditions. Moreover, it has been experimentally proved that the temperature model proposed in the present invention can reduce the error of the accuracy of the estimated temperature by using the temperature model proposed by the present invention compared with the existing temperature model (for example, as shown in FIG. 1). It is within plus or minus 0.8 ° C, and the calculation speed is at least 5 times that of the prior art, and the memory size required to be used is also 0.45% or less of the prior art.
雖然本發明已以實施例揭露如上，然其並非用以限定本發明，任何所屬技術領域中具有通常知識者，在不脫離本發明的精神和範圍內，當可作些許的更動與潤飾，故本發明的保護範圍當視後附的申請專利範圍所界定者為準。 Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and any one of ordinary skill in the art can make some changes and refinements without departing from the spirit and scope of the present invention. The scope of the invention is defined by the scope of the appended claims.
S201~S204‧‧‧溫度推算方法的流程步驟 S201~S204‧‧‧Process steps of temperature estimation method
Claims (14)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW103125528A TWI569136B (en) | 2014-07-25 | 2014-07-25 | Temperature estimating method of microprocessor and computer system using the same |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW103125528A TWI569136B (en) | 2014-07-25 | 2014-07-25 | Temperature estimating method of microprocessor and computer system using the same |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TW201604686A TW201604686A (en) | 2016-02-01 |
| TWI569136B true TWI569136B (en) | 2017-02-01 |
Family
ID=55809643
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW103125528A TWI569136B (en) | 2014-07-25 | 2014-07-25 | Temperature estimating method of microprocessor and computer system using the same |
Country Status (1)
| Country | Link |
|---|---|
| TW (1) | TWI569136B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5513361A (en) * | 1994-07-25 | 1996-04-30 | Intel Corporation | Method and apparatus for reducing power consumption of a fan in a computer system |
| CN2514399Y (en) * | 2001-11-12 | 2002-10-02 | 宝陆科技有限公司 | Temp differential radiator |
| US20110221324A1 (en) * | 2010-03-13 | 2011-09-15 | Zhongshan Weiqiang Technology Co., Ltd | Light emitting diode (led) lamp |
| CN102760115A (en) * | 2012-06-08 | 2012-10-31 | 上海交通大学 | Non-uniform sampling thermal reconstruction method based on dynamic Voronoi picture and device |
-
2014
- 2014-07-25 TW TW103125528A patent/TWI569136B/en active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5513361A (en) * | 1994-07-25 | 1996-04-30 | Intel Corporation | Method and apparatus for reducing power consumption of a fan in a computer system |
| CN2514399Y (en) * | 2001-11-12 | 2002-10-02 | 宝陆科技有限公司 | Temp differential radiator |
| US20110221324A1 (en) * | 2010-03-13 | 2011-09-15 | Zhongshan Weiqiang Technology Co., Ltd | Light emitting diode (led) lamp |
| CN102760115A (en) * | 2012-06-08 | 2012-10-31 | 上海交通大学 | Non-uniform sampling thermal reconstruction method based on dynamic Voronoi picture and device |
Also Published As
| Publication number | Publication date |
|---|---|
| TW201604686A (en) | 2016-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20150220672A1 (en) | Method and apparatus for modelling power consumption of integrated circuit | |
| CN104535257B (en) | A kind of silicon pressure drag temperature-compensating appraisal procedure | |
| Liu et al. | Compact thermal modeling for packaged microprocessor design with practical power maps | |
| Wei et al. | High-precision smart calibration system for temperature sensors | |
| CN111026603B (en) | On-chip network temperature prediction method and device, equipment and storage medium | |
| Levant et al. | Accuracy of some popular non-homogeneous 2-sliding modes | |
| Liu et al. | Multirate multisensor distributed data fusion algorithm for state estimation with cross-correlated noises | |
| Wang et al. | Full-chip runtime error-tolerant thermal estimation and prediction for practical thermal management | |
| CN104950261B (en) | The hardware-in-the-loop test method and system of battery | |
| TWI569136B (en) | Temperature estimating method of microprocessor and computer system using the same | |
| WO2018081945A1 (en) | Method for detecting background noise of sensor, and device thereof | |
| CN109270455B (en) | Induction motor state monitoring method based on weak-sensitivity ensemble Kalman filtering | |
| Sultan et al. | Variability-aware thermal simulation using CNNs | |
| CN106650009B (en) | Method for estimating rapid transient heat distribution of microprocessor | |
| CN114441848A (en) | Power consumption determination method and device, electronic equipment and storage medium | |
| Tan et al. | Fault detection of mechanical systems with inherent backlash | |
| TWI544318B (en) | Thermal prediction system and method thereof | |
| Li et al. | Real-time Ambient Temperature Estimation of Mobile Electronic Devices | |
| CN103186711B (en) | Based on software cost appraisal procedure under non-orthogonal coordinate system | |
| JP2013247023A (en) | Temperature estimation device and temperature estimation method | |
| WO2018045653A1 (en) | Quasi-inverse system control method for nonlinear system | |
| Migórski et al. | An inverse coefficient problem for a parabolic hemivariational inequality | |
| TWI620082B (en) | Behavioral simulation method and system for temperature-sensing circuit design with a digital output | |
| WO2014075361A1 (en) | Voltage detection apparatus and method | |
| Wang et al. | Equivalent load identification algorithm based on least-squares in frequency domain |