Thermal Faraday shield--effectively infinite rejection of thermal forcing

Hi, All,

In my other thread on the power law circuit, I mentioned the thermal Faraday shield, which may be of interest. (Long-but-worthwhile post warning)

Simon and I filed a patent last month on a scheme to improve temperature control by a lot.

*Temperature Control is Slow* Loops controlling the temperatures of macroscopic objects are really slow. The slew rate is slow because the available heating or cooling power can't move the thermal mass very fast. More fundamentally, the bandwidth is limited because thermal diffusion is exponentially slow--asymptotically you get another radian of phase shift for every 1/e worth of rolloff. That makes the usual speedup tricks useless.

• However, it's possible to eliminate that delay by combining the heater and temperature sensor in a single metal element, such as a bit of copper flex circuit. This has been done N times before, but apparently nobody noticed one key fact: If you measure the temperature using the heater drive current, *there's no diffusion delay at all*.

This first came up when I was doing waveguide antenna-coupled Ni-NiO-Ni tunnel junction infrared detectors at IBM, twenty-odd years ago. Unlike photodiodes, low-barrier TJs work by actually rectifying light, so these were basically crystal radios running at 1.6 um. The devices were about a micron across, made by directional evaporation of gold over nickel, with a short Ni-NiO-Ni junction at the vertex. The TJ formed a plasmonic traveling wave structure, so that the ~30 fs RC time constant of the Ni-NiO-Ni system didn't trash the response at 190 THz (1.6 um).

It happened by accident during testing. This plot

is what a reasonably decent device produced in response to a 30-ps pulse at 2.4 um. (I had this very swoopy tunable laser/optical parametric oscillator system back then.)

This second plot

shows what happened when one of my tunnel junctions shorted out during testing. I had about 100 mV of DC bias on it, so even at 3000 ppm/K, a 2.5 mV step is big--a good 8 degrees C. (In this plot you have to mentally subtract the baseline.) The bolometric response looked like a step function on this scale, because it took about 5 us to cool back down, even with 3D heat conduction and 1-um size.

Since the DC current path was about the same as the AC, the near-instantaneous heating of the device produced a near-instantaneous RTD response: about 40 picoseconds.

Applying this idea to normal life, in principle your temperature controller can have any bandwidth you want. Of course the slowness of thermal diffusion means that at sufficiently high frequency the temperature of the RTD decouples from the rest of the world. However, if you tile some surface with these things, you can effectively make a thermal version of a Faraday shield--the huge control bandwidth gives you arbitrarily good rejection of thermal forcing, with no bulky insulation, stirred fluid baths, or big thermal masses.

The decoupling region actually has some interesting features--as the frequency goes up, the amount of material you have to heat goes down, so there's a region where the phase shift is 45 degrees instead of the 90 degrees you get in the low frequency (thermal mass) limit. (This is discussed in Section 20.3 of my third edition,

.)The decoupling also means that in principle the loop bandwidth and compensation don't need to depend on what the element is stuck onto--you get the same huge forcing rejection regardless. You don't even have to worry about windup, despite the slew rate being slow for the bandwidth.

There are a number of control schemes for this, of which my favorite is analog PWM. The heater is in a resistive bridge with a shunt resistor and a reference divider. It gets turned on for a microsecond or so at the beginning by a strobe pulse and an RS-flipflop controlling an NMOS switch, with the . A low-noise amplifier (ADA4899-ish) driving a comparator resets the flipflop when the instantaneous temperature error crosses zero. (The FF is a NAND type, so the heater is turned on if both SET and RESET are active.) It's cool to watch the duty cycle change instantly if you touch the element, and of course the effective loop bandwidth is huge--a short transient gets nulled out in the very next clock cycle.

The Class-H thing I talked about in the other thread is for things like DWDM lasers and OCXOs, where you don't want a lot of EMI right in the sensitive region. (At low power, you can just use an analog loop with a fixed supply.)

There are a whole lot of things you can do with this general scheme, from improved thermolelectric coolers to such things as a battery calorimeter made of metallized mylar like a chip bag.

Fun stuff--suggestions for applications and further enhancements welcome!

Cheers

Phil Hobbs