# Thermal stabilization of TH LED (layout recommendations)

I have to develop a circuit for LED-based optical reference. Due to specific requirements, only available LEDs are in TO-18 packages with metal cans.

The problem is that LEDs experience spectral shift based on die temperature. I would like to avoid this shift by controlling the temperature. So, the question is - how do I improve heat transfer between the die and the heat exchanger?

I plan to have large pads for anode/cathode with thermal vias to the opposite side of the PCB, which attaches to the Peltier element. Measurement thermistor thermally attached to the anode pad on component side, right next to the pin. I was thinking of thermally coupling the can itself as well. Is there anything else I can do about it, perhaps any other recommendations?

UPDATED:

For all the requesters, particular LED "datasheet": https://www.thorlabs.com/drawings/959aa0deecfcd23f-34F04596-F0E3-080D-BD90C82F6B8EC899/LED910L-SpecSheet.pdf

• Please provide a link for the LED you intend to use. Please also state what shift in optical wavelength is acceptable as a working tolerance. – Andy aka Jul 13 '18 at 10:05
• I would consider making a heatsink with holes that fit the TO-18 bodies. – Jasen Jul 13 '18 at 10:55
• What do you mean by "the heat exchanger"? What is this and where is it located with respect to the LEDs? What temperature do you want the LEDs to be? How precisely do you want to control the die temperature? Is there a free flow of air by convection or fan? – Elliot Alderson Jul 13 '18 at 11:22
• You must provide a part number or a link to the datasheet. In the datasheet there should be mounting instructions or a reference to another document with instructions. Some have a thermal base plate and some use the cathode. How much PCB area is available top and bottom? If you have a Peltier selected, will need datasheet for it too. – Misunderstood Jul 13 '18 at 17:55
• Do you know the thermal resistance of standard PCB foil (1 ounce per square foot) is 70 degrees Centigrade per square, of ANY size square? What about the thermal resistance of epoxy fiberglass? (its about 200X that of copper, if I recall rightly). – analogsystemsrf Jul 15 '18 at 3:48

I need a datasheet to give specific recommendations.

I have run many thermal experiments with LEDs in the past couple of years.

First off you should understand that heat flows from warmer to cooler. You must first provide a thermal path from the LED to the peltier plate. This is best done on the component side. Thermal vias should be avoided if there is area on the component side.

There must be a thermal pad on the PCB that attaches to the case or lead(s) of the LED.

This PCB thermal pad must be as large as reasonably possible.

The distance between the LED and peltier plate must be as short as possible.

PCB copper thickness minimum 2 oz.

This is a TO-18 case with a thermal base plate.

This would be my recommended PCB footprint
for the above case with a 40 mm² peltier plate

UPDATE
Now have a datasheet

According to your idea, Peltier is mounted next to the LED body and heat exchanger (radiator, whatever thermal mass) goes on top of that. Thus it is likely, that the field-of-view is covered.

Your half view angle is 12° and the top of the case height and width = 4.7 mm. Using simple right angle trig, the height of a device placed 2.5 mm from the center (0.15 mm from side of LED) of the LED would need to have a height that extended 11.4 mm over the top of the LED to interfere with the beam. 11.4 mm + 4.7 mm LED height = 16.1 mm (0.63") total height.

You need a bar of aluminum or copper to mount and apply pressure to the peltier plate to the PCB. You may as well use a heatsink as the mounting plate with a base less than 16mm - 3.5mm (height of peltier) = 12 mm and anything over 12 mm grind the ends of the fins at a 12° angle.

Then use a low profile fan like the Sunon MagLev HA series with mounted on the heatsink far enough away not to interfere with the LED beam.

2" wide x 2" deep x 0.5" height
Heatsink USA 2" extrusion

Why not do thermal vias, similar to the way it's done with some ICs with exposed pads? Route the heat to the opposite side of the board through vias and remove it there? I'll make some drawings of these ideas and post them tomorrow

Conductive thermal resistance for thermal vias is much higher than a copper plate (PCB layer). See Table 1 and read Sec 3.2 in TI App Note 2020: Thermal Design By Insight, Not Hindsight

Then when the thermal flux reaches the other side of the PCB then you still need the copper plate to spread the heat.

This is how well I found thermal vias to work with high power LEDs:

Now that I have seen the datasheet and your 10 mW LED only generates 0.06 watts of heat. 1.4 Vf x 0.05 A = 0.07 W - 0.01 Optical radiant watts = 0.06 thermal watts.

If you add copper area to dissipate heat, it is likely to generate more of a problem than be of help. I doubt you will be able to determine if the LED is on of off by touching the TO-18 case. This means there is very little temperature difference Δ between the case and ambient. With a small temperature Δ there will be less thermal conductivity and convection and it will take longer to reach thermal equilibrium. So you would have to wait longer (hours) to use the LED.

1. As the temperature changes Vf changes.
2. As Vf changes the amount of heat generated changes.
3. As the amount of heat generated changes the temperature changes.
4. Go to step 1.

This is why the datasheet has no thermal information for a 10 mW LED.

I actually expect the ambient (outside the device) temperature to be within -30'C to +40'C (outdoors usage in whole Europe from Italy to Norway/Finland all year round)

Oh!

ThorLabs offers up to 100mW electrical

Good. You will need a heater as well as a cooler.
It is much easier to heat on a cold day than cool on a warm day.

A temperature sensor would be a very good thing to have also.
Two temperature sensors (ambient and LEDcase) would be good to have.

Then you can use a micro-controller to turn peltier plate off & on and adjust the fan speed based on the case sensor.

Use the ambient sensor to choose the methodology (e.g heat or cool). At a specific ambient temperature the peltier and fan will net be needed. A case temperature only sensor will not give enough information.

The question is can you lower the junction temperature on a hot summer day to the same temperature as a cold winter day? If not, you have to add another heat source. You could attach the heat source to the opposite side of the PCB and use thermal vias. If needed, I would recommend using an inefficient, low efficacy, CoB LED as a heat source with a dimable LED driver (e.g. TI LM3414). You may even be able to use a thermal foldback circuit with the driver to turn the CoB on when it gets too cold. The LM3414 has a thermal foldback but you'd need a different thermistor.

IMPORTANT

I forgot to mention this previously. Do NOT use HASL plating on the PCB, use ENIG. HASL is much too uneven for an effective thermal interface. ENIG has very good planar characteristics (therefore thermal interface) .

You might consider a small board for just the LED. Qty 5 PCB with 4 oz copper, ENIG plating, 0.4 mm thickness (thin PCB if thermal vias are needed 1), 100mm x 75mm costs $171 for qty 5 at PCB Way,$5 ea at qty = 100.

1Reducing PCB thickness by 50% (1.6mm => 0.8 mm increases thermal conductance of vias by 2X. 1.6 mm => 0.4 mm = 4X improvement.

• Edited the post to include "datasheet". Sadly, it does not contain anything about mounting/thermal characteristics. According to your idea, Peltier is mounted next to the LED body and heat exchanger (radiator, whatever thermal mass) goes on top of that. Thus it is likely, that the field-of-view is covered. Why not do thermal vias, similar to the way it's done with some ICs with exposed pads? Route the heat to the opposite side of the board through vias and remove it there? I'll make some drawings of these ideas and post them tomorrow – stiebrs Jul 15 '18 at 17:12
• Thank you for healthy discussion. I actually expect the ambient (outside the device) temperature to be within -30'C to +40'C (outdoors usage in whole Europe from Italy to Norway/Finland all year round), so the temperature gradient might be quite considerable and I want to decouple LED from it. I don't mind 30 minute thermal stabilization time. I do plan to create a thermal chain LED-PCB-copper-paltier-radiator-outdoors convection. As you noted, this is not a high power led (ThorLabs offers up to 100mW electrical), so it shouldn't need massive amounts of heat conducted to/from. – stiebrs Jul 17 '18 at 6:32
• I wasn't quite done when you commented. Then I added some more to address this comment. – Misunderstood Jul 17 '18 at 7:38
• Is the 100 mW optical or electrical. Their 10 mW is optical radiant watts and the electrical (Vf x current) is 70 mW. – Misunderstood Jul 17 '18 at 7:53

You can just drill a hole in a piece of aluminum or copper and put the LED in place with heatsink compound. Considering the internal construction of the package you may wish to use a carrier PCB with thick copper mated to the back of heatsink, again with some thermal compound to fill the voids.

If you really care about the temperature you can use a TEC to control the temperature (or just use a heater and elevate it above ambient). This method is used to stabilized the wavelength of some laser diodes for spectroscopy.