Abstract:
Starting from 2300 transistors on a chip in 1971 by INTEL to 192 billion transistors on a single chip by AMD in 2017, the technology has grown faster than expected and so did the leakage currents and power consumption on these chips. These modern systems tend to have a behavior similar to positive feedback i.e. leakage currents increase heat dissipation, increasing the temperature of the chip which in turn exponentially increases leakage current. In a recent report it was reported by intel [1], that for every 15 °C rise in temperature delay will be increased by approximately 10% - 15%.
On-chip temperature monitoring has become inevitable in the current scenario when chips are running at exceptionally high frequencies dissipating lots of heat to chip vicinity, which creates performance and reliability issues in the microprocessor. In DRAMs, the self-refresh rate is dependent upon current die temperature. On-chip temperature sensors provide critical feedback to the dynamic voltage and frequency scaling blocks that are implemented to prevent excessive chip heating which can destroy the device or reduce the expected lifetime. Multiple small temperature sensors are distributed throughout the chip at expected hotspots to monitor on die temperature, in applications where excessive heating of chip is of concern
