editor's blog
Subscribe Now

Taking the Temperature

Three of this year’s ISSCC’s sensor papers related to temperature sensors, although with different approaches and goals.

The first of them addresses the needs of tough environments such as those that automotive and military applications require. Their main motivator was the fact that, at high temperatures, traditional on-chip bipolar junction transistor (BJT) temperature sensors become much less accurate, so designers end up opting for off-chip thermistors and such.

So their idea was not to measure temperature through transistor characteristics, but rather to measure the thermal diffusivity – how fast heat travels, which is proportional to the -1.8 power of the absolute temperature. Measuring the diffusivity is conceptually easy: have a heat source (a resistor) and measure how long the heat takes to travel through the silicon to a nearby thermopile.

Rather than just measuring time, however, the resistor is pulsed and the response is detected as a low-frequency signal. When the temperature changes, it takes more or less time for the signal to reach the sensor, resulting in a phase shift in the signal.

The challenge here is that the frequency needs to be accurate, requiring an accurate time base. To avoid that, they did a second such sensor as a reference, this time including an oxide trench between the resistor and sensor. Now the heat has to go through the oxide en route, and the oxide thermal diffusion is about 100 times less that than of silicon, and is about 20 times less dependent on the absolute temperature. Using this as a reference, they were able to cancel out variations.

Combined with a factory room-temperature trim that took care of trench spread variation, they achieved a 0.4 °C accuracy.

They noted that SOI works better for this because the underlying oxide lets less of the heat leak away, making the whole thing more sensitive.

They did acknowledge that power is a bit of an issue for this approach…

You can find more detail on this in Session 11.5 of the proceedings.

Meanwhile, two other on-chip sensors relied on measuring ratios to nail the temperature. The first was intended for use in RFIDs, allowing temperature measurements to be sensed by an RFID reader. It therefore needed to be quick and energy efficient.

Their approach measures the ratio of the VBE of a p-type BJT against the difference in the VBEs of two different p-type BJTs. VBE goes down as temperature increases, giving it the cryptic-sounding name complementary-to-absolute-temperature, or CTAT. The VBE difference, however, goes up with rising temperature, making it proportional-to-absolute-temperature, or PTAT.

The ratio is known to vary from 6 to 28 over the military temperature range, so measuring that ratio lets them figure out the temperature. They used a two-step process here, with successive approximation (SAR) giving them a coarse reading, followed by a “zoom” ΔΣ ADC.

Their resulting accuracy was 0.15 °C. The circuit draws 3.4 µA, higher than some other works, but it operates very quickly, providing a lower energy-per-conversion (as a figure of merit) compared to prior works.

You can get more detail on this work in Session 11.7 of the proceedings.

The last BJT ratiometric approach was presented by Intel. Their use model is different, since they’re trying to measure various hot spots on the chip so that they can throttle or even shut things down if they get too hot. The problem is that the exact location of the hot spots is typically not known until very late in the process – possibly even after first silicon, so the sensors must be small enough to insert and move around at the last minute.

They also need them to be fast. They refer to the need to measure “gradients,” which highlights the fact that “gradient” has two meanings. There’s a lot of discussion these days about the effects of temperature gradients across a die – here we refer to the fact that two different points have different temperatures, and that there’s a gradient in the silicon between them.

However, the Intel paper uses “gradient” to refer to the dynamic change in temperature over time. Because they want to be able to detect sharp rises in temperature, the sensor has to work quickly enough to identify such spikes. They can, however, tolerate more error: ±1 °C near throttle temperatures, and even more at lower temperatures.

The ratio they chose was different from that of the prior paper: they used a bandgap reference to generate a CTAT VBE and a temperature-independent reference voltage. They turned the voltages into frequencies that drove counters; the ratio was available from the counters. Further chopping was done to eliminate mismatch errors and noise, taking the noise level down from 0.73 °C to 0.19 °C.

More information on this paper can be found in Session 11.8 of the proceedings.

Leave a Reply

featured blogs
Oct 23, 2020
Processing a component onto a PCB used to be fairly straightforward. Through hole products, a single or double row surface mount with a larger center-line rarely offer unique challenges obtaining a proper solder joint. However, as electronics continue to get smaller and conne...
Oct 23, 2020
[From the last episode: We noted that some inventions, like in-memory compute, aren'€™t intuitive, being driven instead by the math.] We have one more addition to add to our in-memory compute system. Remember that, when we use a regular memory, what goes in is an address '...
Oct 23, 2020
Any suggestions for a 4x4 keypad in which the keys aren'€™t wobbly and you don'€™t have to strike a key dead center for it to make contact?...
Oct 23, 2020
At 11:10am Korean time this morning, Cadence's Elias Fallon delivered one of the keynotes at ISOCC (International System On Chip Conference). It was titled EDA and Machine Learning: The Next Leap... [[ Click on the title to access the full blog on the Cadence Community ...

featured video

Demo: Inuitive NU4000 SoC with ARC EV Processor Running SLAM and CNN

Sponsored by Synopsys

Autonomous vehicles, robotics, augmented and virtual reality all require simultaneous localization and mapping (SLAM) to build a map of the surroundings. Combining SLAM with a neural network engine adds intelligence, allowing the system to identify objects and make decisions. In this demo, Synopsys ARC EV processor’s vision engine (VPU) accelerates KudanSLAM algorithms by up to 40% while running object detection on its CNN engine.

Click here for more information about DesignWare ARC EV Processors for Embedded Vision

featured paper

An engineer’s guide to autonomous and collaborative industrial robots

Sponsored by Texas Instruments

As robots are becoming more commonplace in factories, it is important that they become more intelligent, autonomous, safer and efficient. All of this is enabled with precise motor control, advanced sensing technologies and processing at the edge, all with robust real-time communication. In our e-book, an engineer’s guide to industrial robots, we take an in-depth look at the key technologies used in various robotic applications.

Click here to download the e-book

Featured Chalk Talk

SLX FPGA: Accelerate the Journey from C/C++ to FPGA

Sponsored by Silexica

High-level synthesis (HLS) brings incredible power to FPGA design. But harnessing the full power of HLS with FPGAs can be difficult even for the most experienced engineering teams. In this episode of Chalk Talk, Amelia Dalton chats with Jordon Inkeles of Silexica about using the SLX FPGA tool to truly harness the power of HLS with FPGAs, getting better results faster - regardless of whether you are approaching from the hardware or software domain.

More information about SLX FPGA