How To Reduce Thermal Guard-Banding

Accuracy in temperature sensors can have a big impact in designs from 40nm down to 7nm and beyond, reducing the amount of guard-banding that is required, which in turn can lower the power and extend the life and reliability of components. But at these process geometries, not all sensors measure temperature equally.

Thermal guard-banding is a very important consideration for design teams, and the best way to reduce guard-banding is through embedded in-chip temperature sensors. Here’s why. Consider two different temperatures sensors, both un-calibrated, one slightly more accurate, one slightly less accurate. There is a scale of die temperature, and a target temperature of 85°C. The software could be used to slow down clock frequencies in order to bring the temperature of the device down, or maybe to set a thermal/temperature alert within the software.

If you have a target temperature of 85°C and your temperature sensor is, say, +/- 5°C accurate, then you have a set point or a range of temperatures that can vary between 90°C and 80°C. Your software will need to be set at its lowest point (worst case point) to 80°C. Then, allowing for the inaccuracy of the temperature sensor, we are compounding this issue. We still have to allow for the +/- 5°C, so then the lowest point actually becomes 75°C. Now, if we take a more accurate temperature sensor with an accuracy of +/-2°C, again un-calibrated, you then have a temperature range of 87°C and the lower part of the range being 83°C. If you are setting your software to take action at that level, then you still have to take into account the inaccuracy, which is +/-2°C. So that comes down to 81°C

The important point is we are now comparing the lowest set point for software at 81°C with a good temperature sensor, and for a less accurate temperature sensor it is 75°C. Just by using a slightly better temperature sensor in this case you could be saving yourself 6°C of die temperature.

What does that mean for the customer? Depending on architecture and application, 6°C better accuracy could save anywhere between 5 and 10 watts of power. That, in turn, extends the battery lifetime for applications such as IoT and consumer devices. For data center or telecoms applications that can be used to optimize speed and throughput of the device, and for automotive it can increase the reliability of the devices because you are then able to help mitigate issues to do with stress and electro-migration on the devices.

Accurate thermal sensing is a critical part of in-chip monitoring subsystem solutions.