Digital Temperature Sensing in a Variable Supply Environment EE241 Term Project Matthew Spencer Steven Callender Spring 2009.

Digital Temperature Sensing in a Variable Supply Environment

EE241 Term ProjectMatthew SpencerSteven Callender

Spring 2009

Classic Temperature Measurement is Analog

• Accurately Measures BJT Current • Big, Power-Hungry, Analog Circuit

[1] M. Pertijs, K. A. Makinwa, J. H. Huijsing. A CMOS Smart Temperature Sensor With a 3σ Inaccuracy of +/-0.1°C From -55°C to 125°C. IEEE JSSC, December 2005

Digital Replacements Ignore Real Effects

[2] Poki Chen et al. An Accurate CMOS Time-to-Digital-Converter-Based Smart Temperature Sensor. JSSC 2005

• Measure Delay Pretty Well• Ignore Supply and Process

[3] K. Woo, D. Ham et al. Dual-DLL-Based CMOS All-Digital Temperature Sensor for Microprocessor Thermal Monitoring. ISSCC 2009.

Outline

• Motivation

• Current sensors: What makes them

good?

• Comparing digital sensors and

Simulation

• Digitally Assisted Supply Correction

• Supply Insensitive Biasing

A Worrisome Graph from a Paper …

• Vdd to delay constants mismatched• Optimal Resolution, not supply rejection

“Linear With Supply” Costs a Lot of Area

Can the We Correct for Supply with Math?

• Use Digital Algorithm to Help Measurement• Use Several f(C) to Guess Vdd and T

Digital Correction Can’t Handle Vdd Errors

• Good Vdd Guess, but Bad T Guess• Exponentially sensitive to Vdd

Accounting for Vdd shifts: A Better Approach

• Strong coupling between Vdd and temperature in delay variations• Very hard to isolate each effect for a given delay variation (How much is due to Vdd and how much is due to temperature?)

• Solution: Supply insensitive biasing• Idea: “remove” any Vdd shifts from the point of view of the delay cells

Accounting for Vdd shifts: A Better Approach

• Once delay variations are decoupled from supply shifts, problem becomes trivial

• Supply insensitive biasing is the most viable approach, but there are still issues:• Device Sizes• Topology: New approaches may need to be explored

Conclusion

• Literature ignores real environments

• Correcting for supply hasn’t worked yet

• Delay is a BAD temperature metric

• … But finding something else is hard

Thank You

Questions?