Compact Thermal Sensors in Intel Processors from 90nm to 22nm

February 17, 2014
Location: 825 Mudd
Hosted by: Peter Kinget
Speaker: Dr. Joseph Shor, Intel


Temperature sensing and regulation in microprocessors are important for power management and have a direct influence on power/performance. When the chip gets too hot, its frequency gets lowered in a process called throttling, in order to prevent it from exceeding the maximum reliable operating point. As temperature is lowered, the frequency of the chip can degrade because of the ITD effect (Inverse Temperature Dependence). In this case, information from the thermal sensor is used to raise the supply voltage to maintain performance. The thermal sensor is also used for fan regulation and prevents the chip from burning itself up in thermal runaway. In the older chips, there was a single core, so it was sufficient to have one or two thermal sensors per IC, and the power and area were not very important. Since the industry shifted to multicore, there can be as many as 25 sensors per chip, so their area and power have become substantial. In addition, analog circuits tend to have very poor scaling between process generations, or even reverse scale. In the Nehalem generation the thermal sensors occupied more than 1% of the overall die area, and consumed > 200mW, so there was much motivation to scale them down. In our group in Yakum, Israel, we have developed compact and power-efficient thermal sensors. Using novel analog architectures and circuit techniques the area and power were scaled by > 10X in the same process generation. The level of scaling in analog is virtually unheard of. In the Sandy-Bridge processor, we implemented an NMOS based sensor, which measured the temperature dependence of threshold voltage (Vt) and mobility (Mu) and output a temperature-dependent frequency. In addition, we also developed a very compact diode-based thermal sensor which used a band-gap reference to generate a temperature-dependent voltage, which was measured using a frequency-based analog-to-digital converter. The diode sensor is considered more reliable than the NMOS based design since Vt and Mu change during the process evolution. The 22nm sensor was one of the world's first analog circuits reported at this process node. These sensors have been benchmarked as the smallest IC thermal sensors in the art, with sizes < 0.006 mm2. They also have a very fast sampling speed (> 5kS/sec), which enable them to capture fast thermal transients on the processor. The designs are very robust and can be manufactured with > 99.9% yield. The evolution of these sensors from 90nm down to 22nm will be discussed, along with some of challenges inherent in designing analog circuits in the noisy, pure-digital environment of a microprocessor.

Speaker Bio

Joseph ShorDr. Joseph Shor received PhD in Electrical Engineering from Columbia University, 1993. BA in Physics, Queens College, 1986. Dr. Joseph Shor has published more than 50 papers in refereed Journals and Conference Proceedings in the areas of Analog Circuit Design and Device Physics. He holds 35 issued patents and several pending patents. Since 2004, he has been at Intel Corporation, where he is presently a Principal Engineer, and head of the Analog Team at Intel Yakum in Israel. Between 1999-2004, he worked at Saifun Semiconductor as a Staff Engineer where he established the analog activities for Flash and EEPROM NROM memories. From 1994-1999, he was a Senior Analog Designer at Motorola Semiconductor in the DSP Division. From 1988-1994, he was a Senior Research Scientist at Kulite Semiconductor, where he developed processes and devices for Silicon Carbide and Diamond Microsensors. His present interests include Switching and Linear Voltage Regulators, thermal Sensors, reference circuits, PLLs , IO circuits, and low voltage/power analog circuits. He is also interested in microsensor design, physics and applications. He is an IEEE Senior Member and a member of the ISSCC Technical Program committee.

500 W. 120th St., Mudd 1310, New York, NY 10027    212-854-3105               
©2014 Columbia University