Precision ICs Need to Take Temperature Measurements Into Account

Thermal issues abound as electronic integration increases density of integrated circuits. In fact, this was the topic of a recent Planet Analog-Integration Nation live chat: How can we best get the heat out? In that discussion, a main issue discussed was increased thermal loading in denser integration leading to damaging, component-life-shortening temperature levels.

Work on measuring temperature in ICs goes back a long way. For example, in 1977 Allegro Microsystems published an application note entitled “Computing IC Temperature Rise.” The note gives basic equations to calculate junction temperature rise as a function of power, with some duty cycle considerations. In cases where that approach is inadequate, Allegro suggested using the forward voltage vs. temperature relationship of a diode within the IC as a direct measurement. Of course, that approach requires advance planning in the design and adds cost to the IC for the sense diodes.

Almost 35 years later a TI-authored paper in the Electronic Engineering Journal, “Methods of Estimating Component Temperatures,” promoted exactly that technique as the method of choice. The authors also talked about other ways to estimate IC temperature, ranging from measuring the board temperature near an IC to measuring temperature on the surface of the package. These methods have tradeoffs and are less accurate than in situ measurement.

There is also a long history of bench-level and laboratory measurements of IC temperature. A 2006 paper from the Institut National des Sciences Appliquees (INSA) de Lyon in France introduced scanning thermal microscopy as an analytical tool to study temperature rise and distribution in IC packages. Although capable of very accurate measurements, it obviously isn't something every development group will have available.

Upping the ante, in 2009 a team from the UC Santa Cruz EE department described the use of confocal Raman spectroscopy as a way to acquire 3D temperature profiles of a functioning IC package. By appropriate choice of light source for the Raman Spectroscopy, they claimed to measure 3D positions of heat sources, using a specially fabricated device as the test reference. My take is that this method might be of limited value in real ICs where there are a lot of heat sources.

With all this focus on heat in ICs, it is almost ironic that in the area of silicon photonics, researchers are developing methods of heating the chips. In a new paper from the silicon photonics leaders at IMEC and Ghent University, architectures for heating elements built into an IC are compared. What the researchers showed is that using existing CMOS processes and element compositions, a large range of performance curves can be obtained, which gives chip designers a lot of design freedom.

The reason that heating elements may be needed in silicon photonic ICs is that silicon waveguides have a significant shift in optical behavior with temperature. In WDM (wavelength-division multiplexing) systems, which are key to optical networking performance, wavelength control is paramount. The authors of the IMEC/Ghent study state that silicon waveguides can cause shifts of 80ppm/°K. Considering an IC might need to work over a span of 125° or more, this can cause shifts of 10 nanometers. Since WDM systems use channel spacing of less than 1 nm, it is evident that temperature compensation is needed.

In order to do very precise temperature compensation, you need good temperature measurements. This brings us full circle.

Interferometry has been used in many ways to perform temperature measurements, because it allows detection of very small, mechanical shifts associated with the thermal expansion of materials. In fact, in the IMEC/Ghent study described earlier, they used a Mach Zender interferometer to measure temperatures in the heating element evaluation. In a new paper from the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, and A*STAR (Agency for Science, Technology and Research), Singapore, the authors describe measuring temperature with a Michelson Interferometer constructed in CMOS1 that is only 40μm x 70μm and uses silicon waveguides. You might recall the Michelson interferometer from the famous Michelson-Morely experiment, which showed there was no “aether wind” and eventually led to Einstein's Special Relativity. This new instance is much, much smaller!

Called a photonic thermocouple by the authors, the work is directed at low-cost, high-sensitivity, and high-temperature-range applications, especially in harsh environments. What caught my attention is that this device is ideally suited for integration into silicon photonic devices. Figure 1 is taken from the paper.

Figure 1

SEM image of the silicon waveguide geometry used to form a Michelson Interferometer. The loops act as 'mirrors' causing full reflection, and the section in detail 'DC' is a splitter.(Source: Reference 1)

SEM image of the silicon waveguide geometry used to form a Michelson Interferometer. The loops act as “mirrors” causing full reflection, and the section in detail “DC” is a splitter.
(Source: Reference 1)

The authors show that their device can produce a linear response over at least 100K° of temperature change, and with multiple times the sensitivity of other methods such as the temperature-dependent response of a silicon waveguide ring resonator. Although this work wasn't directed at silicon photonics integration, I think it may have a useful place in the complete solution of silicon photonic ICs. What do you think?


  1. Photonic Thermocouple Design Based on an Ultra-Compact Michelson Interferometer, by J. F. Tao, H. Cai, Q. X. Zhang, J. M. Tsai, P. Kropelnicki, A. B. Randles, M. Tang, and A. Q. Liu. The paper was published by the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, and A*STAR (Agency for Science, Technology and Research), Singapore.

Related posts:

18 comments on “Precision ICs Need to Take Temperature Measurements Into Account

  1. eafpres
    September 24, 2013

    In the article, the statement 80 ppm/°K should say 80 pm/°K.  That means a 125K temperature range will cause a length change on the order of 10 nm, which is much larger than the channel spacing in DWDM.  Thus, some kind of temperature control or compensation method is necesary.

  2. Davidled
    September 24, 2013

    I am wondering if this type temperature measurement could be useful for the vehicle today. But, in the future, as vehicle is getting smart year after year, this precision sensor might be required. In the auto lab, for testing system, engine development requires a quite precision temperature in the powertrain. I think that it may be a good beneficial for engine controller.

  3. eafpres
    September 24, 2013

    Actually the authors suggest the device is useful for harsh environments. So it might be used in exhaust gas T or even inside the combustion chamber.

  4. samicksha
    September 25, 2013

    would like to have more description and brief for the below graphs..apart from o/p reflection spectra of sensor when the environment temperature changes from 13ºC to 95ºC. (b) Environment temperature versus interference wavelength shift.

  5. Netcrawl
    September 25, 2013

    The works on optical temperature sensors capable of operating in harsh environment is currently a hot topic that has attracted considerable interest such as the automotive and railroad industry where there a great need for effective tools in their engine controller

    Using silicon optical waveguide to sense the temperature is considered as a more effective way due to silicon is a common material and has mature fabrication technology. It also has the advantages of higher cost effective and more sensitivity than fiber-based sensors. 

  6. eafpres
    September 25, 2013

    @samicksha–the authors were monitoring their device with an external spectrophotometer coupled via a waveguide.   The curves in the top figure are the shift in the power transmitted through the interferometer vs. wavelength–it is scanned by the spectrophotometer.  The line is simply the fit of the wavelength of maximum interference (i.e., the point where the interferometer has maximum dextructive interference, which therefore is the minimum transmitted light) vs. temperature, where again the temperature was controlled in their experiment externally.

    To use this device as part of sensing temp in a Silicon Photonic integrated circuit, some method of sensing the light would be needed.  For a sensor, they would presumably package it with a solid-state laser of some sort and a sensor with a wide enough spectral range to detect the transmission minima.

  7. RedDerek
    September 25, 2013

    One interesting application would be use the on-chip temp sensor as a feedback to an on-board heater circuitry. Thus one can use the feedback to keep the IC at a constant temperature to ensure all the silicon is acting within very tight tolerances over a wide range of external temperatures – sub-0 to a hot Saharan desert.

    Not sure of how much a need would be out there to do this requirement.

  8. eafpres
    September 25, 2013

    @RedDerek–That is the application that I was thinking about; the use for it would be in Silicon Photonics–once you integrate a light source and other optical components onto a chip, you will need temperature control.  Most systems therefore operate above ambient temperature and are controlled, as you describe.

  9. samicksha
    September 26, 2013

    Thanx eafpres, but can we use double/ single beam spectrophotometer for photonic IC, as it can help us in comparing the light intensity between two light paths…

  10. Brad_Albing
    September 26, 2013

    @eafpres – looks like an editing error. If I find out who did that, I'll give them a stern talking to.

  11. RichQ
    September 30, 2013

    Perhaps one application of this would be a temperature compensated clock source where you keep the circuit at a known temperature (higher than ambient) so that you minimize drift.

  12. Steve Taranovich
    September 30, 2013

    Aside from on-chip temperature measurement, there is a great paper on modeling temperature of an IC, “A Behavioral and Temperature Measurements-Based
    Modeling of an Operational Amplifier Using VHDL-AMS” by IMS Laboratory and Schlumberger in France.

    Although this technique deals with a temperature-dependent op-amp model and the model is especially developed for high temperatures, I believe it can be expanded to a more complex circuit model as well.

  13. Brad_Albing
    September 30, 2013

    @Rich – yep – I've seen that done before. Sometimes it's easier to elevate and then control (servo) the temp at some point above ambient than to make the ckt work the way you want it to over a broad range of somewhat unpredictable temps.

  14. Brad_Albing
    September 30, 2013

    @Steve – Thanks ST, that's a good one to check out.

  15. Steve Taranovich
    September 30, 2013

    The IEEE XPlore site has a wealth of information and is a system that goeas back to articles and white papers in the early part of the 20th century if you need to

  16. Brad_Albing
    September 30, 2013

    @ST – and some of those papers from back in the early 20th century are a treat to read – when the IEEE was just the IEE. Just “Electrical” – no “Electronic” in their name.

  17. Steve Taranovich
    September 30, 2013

    Hi Brad—Interesting about “electrical” vs. “electronics”—EDN magazine had it's first issue in 1956 and EDN stood for Electrical Design News. The articles were about motors, motor controls, some vacuum tube designs, power electrical engineering, etc.

  18. SunitaT
    October 29, 2013

    The AD592 is a two incurable monolithic IC temperature transducer that offers an output current proportionate to complete temperature. For a wide variety of supply currents the transducer turns as a high impedance temperature dependent existing source of 1 µA/K.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.