January - May 2024
In typical ion trap experiments with calcium, light at 397nm from an ECDL is used for Doppler cooling. With barium, the typical wavelength is 493nm and comes from either a direct 493nm ECDL or a frequency-doubled 986nm ECDL. That is, after the labs stopped using awful dye lasers which in some cases were the only laser technology available at the time - the seminal paper from 1978 by Neuhauser, Hohenstatt, Toschek and Dehmelt used two of these to cool a cloud of barium ions.
In mainland China and Russia, ECDLs are getting replaced with frequency-converted combinations of high-power infrared fiber lasers, which are inefficient, bulky and expensive, but easier to control and stabilize, narrow-linewidth, and less vulnerable to vibrations. They also enable more workarounds against the circus of export control regulations and patents of the past couple years.
While ECDLs are an improvement over dye lasers, they are not a panacea.
Firstly, not all laser diodes behave well with external cavities, and I have yet to find a single ECDL physics paper where the diode that reportedly worked could be purchased, at least without involving a sketchy vendor asking a four-digit USD price for something that is not fundamentally different from the diode inside a $5 pointer. For GaN diodes used at 397nm, it appears to be often required that they be AR-coated or even have a special waveguide shape, which puts them deep into quantum gravy train territory where you are paying through the nose before you even get the first photon.
Secondly, the addition of the grating and its associated precision mechanics greatly increases complexity and cost, and makes the laser sensitive to vibrations. "Cateye" ECDLs are somewhat less vulnerable to vibration, but a suitable interference filter for the target wavelength and the cateye reflector are more difficult to find than gratings, they add more degrees of freedom for things to go wrong, and they are expensive. It's also hit-or-miss whether a filter of a given bandwidth will work with a particular diode, and you have to cough up at the cash register at every attempt.
The excellent W's Laser-Projects page on amateur holography goes in-depth about the trade-offs between an ECDL and a free-running Fabry-Perot diode, with many practical examples. Philip C. D. Hobbs' book also includes very useful information about dealing with free-running diodes (though his stance on running diodes single-frequency at non-infrared wavelengths is "don't do that" :). Don't count on "quantum" laser companies to educate you on these topics, see for example this infographic from Vescent. Their estimate of the linewidth of a laser pointer is off by 4 to 5 orders of magnitude, particularly on the cheaper (with just resistor and switch) red models when the batteries run low. DFB diodes for coherent optical communication systems can have linewidths as low as 50kHz, whereas Vescent says it's above 1MHz. Frequency combs can be built with inexpensive telecom components (note that their $5,000 estimate can be very substantially cut down by sourcing from China whilst avoiding tariffs and taxes). The only clear separations between "Quantum 2.0" lasers and others are their price, unreliability, unavailability, and geopolitical nonsense.
For applications that need precision in optical frequency, the only clear advantage that an ECDL has over a free-running diode is its narrower linewidth. The linewidth is narrowed in an ECDL because the cavity contains an air/vacuum section that is not subject to the Henry effect, and which is much longer than the diode's semiconductor section where Henry linewidth broadening takes place. For a free-running Fabry-Perot diode, we can expect a linewidth on the order of dozens of MHz, which is not ideal for Doppler cooling, but may not be a show-stopper: after all, the Ba+ transition at 493nm has a natural linewidth of 20MHz (source) and the Ca+ transition at 397nm has a natural linewidth of 22MHz (source). So, it seems worth researching whether a carefully controlled free-running diode can be used to reach the Lamb Dicke regime, even if the ions may not end up deep into that regime!
For Doppler cooling, it is highly desirable that the laser operates on a single optical frequency, since additional frequencies make spectroscopy locks more difficult and noisier, increase unwanted scattered light in ion traps, may drive unwanted transitions or cause heating, and can lead to mode partition noise.
Single-frequency operation is also of particular interest to the holographer, as the presence of additional frequencies can easily ruin a hologram. On W's Laser-Projects page, there are detailed maps of the current-temperature plane that indicate where diodes run single-frequency (if they do at all). Those are automated computer scans using a second-hand high-resolution grating spectrometer that was obtained from eBay. Only the complete newbie thinks that searching eBay for similar spectrometers would turn up anything interesting, at least with a chance higher than winning a lottery. The slightly more experienced laserist already knows that, in the unlikely event that a listing for such an instrument would appear, it would be expensive or broken beyond repair or both. In the USA, some zealous LSOs interpret laser disposal regulations as meaning that diffraction gratings and other passive optics must be thoroughly scratched before scrapping any optical device - you can guess how I found out.
The affordable and reproducible way is to collimate the beam and shine it onto a highly dispersive grating (for example a holographic grating at 3600 lines per mm) near grazing incidence, and observe the diffraction pattern after a few meters on a piece of paper taped onto the wall. For near-UV wavelengths such as 397nm, the brightening agents which are an ingredient in most modern papers fluoresce and make the laser light appear bright blue and easily visible.
For a well-behaved diode, as the current is ramped from zero, you will first see a line made of equally spaced dots, which is the diode's ASE filtered through the Fabry-Perot interferometer formed by the diode's internal mirrors. As you keep increasing the current, the diode will reach threshold and one of the dots will start lasing and shine much brighter. This is the single-frequency region that we are interested in. The lasing dot may jump around and flicker depending on the current and temperature. At even higher currents, additional dots will start lasing simultaneously, and mark the end of the single-frequency region.
The SHARP GH05030C2LM 505nm diodes are pretty close to such ideal behavior, and will even run stable single-frequency off a bog standard constant-voltage power supply with a current-limiting resistor and no active temperature stabilization. Unfortunately they cannot be tuned to 493nm even when cooled to very low temperatures - after a point, lowering their temperature just increases the forward voltage and does not change the wavelength anymore. Other diodes may show a more complicated behavior, and would be more or less usable - many examples are on W's page.
Gratings suitable for detailed spectral observation are Thorlabs GH13-36U below 535nm, and Thorlabs GH13-24V above. I have tried a few lower-cost Chinese gratings, but they were of very bad quality and completely unusable. Gratings are one of the few Thorlabs products that are reasonably priced ($99.50 each for these), so it does not sound particularly worthwhile to spend much time on alternatives - at least until politicians come up with an even worse set of export control ideas. Like many scientific suppliers, Thorlabs tends to overcharge international shipping; it helps to have an account with one of the major shipping companies and ask Thorlabs to ship collect. Thorlabs also uses procedures that attract customs attention; therefore, in most parts of the world, local customs will intercept the package and hold it to ransom until they get their pound of flesh, but you probably know the drill.
The comprehensive and automated plots from W's Laser-Projects page are an interesting scientific curiosity and are certainly useful to optimize a laser for maximum margins of operation in a single-frequency region, but you can already get a pretty good idea of whether a diode is well-behaved or not by just playing with the current and temperature manually. Also note that the single-frequency regions always appear at low currents near threshold, so you don't really need to look at higher-current regimes.
For determining the wavelength of lasers with a precision of a few tenths of nanometers, there are a few options. A recent one is the handheld HP320 spectrometer from China. It costs about $380, it is very convenient to carry and use, and it works as advertised with 0.2nm wavelength accuracy. On the negative side, its construction is slightly janky (take a laugh at the mounting of the Bluetooth module), has some silly features (it looks like its designers really wanted to build an iPhone), and the firmware could be improved (USB bugs, no multi-peak detection, no zoom feature on spectrum display). But overall it is a very good deal.
USB spectrometers made by Ocean Optics are common enough that eBay is a somewhat credible source for them. They have a Linux driver which is reliable and pretty bug-free.
If you want to build your own, Les Wright's PySpectrometer looks like a good place to start.
The idea is to be able to control the diode's temperature over a large range (which corresponds to a large wavelength tuning range and better chances of taking a given diode to the required wavelength), and stabilize it (so we can keep the diode single-frequency and at the wavelength we want). We want to cool the diode and not heat it up, since operating diodes at elevated temperatures reduces their lifetime (they can be surprisingly tough - I managed to run a room-temperature 487nm GH04850B2G at 493nm for several hours at a searing 160C, but its output power progressively degraded and the idea of heating diodes was not investigated further).
Condensation on a laser diode is a serious problem - it attracts contaminants which can be very difficult to get rid of. To prevent it, the diode and the other components were placed inside a hermetic IP68-rated aluminum "stomp box" with a few bags of moisture-absorbing silica gel. Those were simply recycled from food items, and dried by placing them on a hot plate at 120C for a few hours. A stack of three TECs was glued to the bottom of the stomp box using heat-conductive epoxy. The laser diode was placed at the top, inside a simple aluminum block with a hole drilled to accomodate common collimator tubes, and the block was glued to the top TEC. On the side of the box, a window was glued on top of a hole to let the beam through but not humidity. Aluminum stomp boxes are very easy to drill, an inexpensive cordless hand-held drill is sufficient. The window should be assembled at an angle to prevent the Fresnel reflection from going back into the laser diode. Thermistors were glued to the aluminum block and between the top and medium TECs. Wires for the TECs, diode and thermistors went through another hole in the box, and the interface was made hermetic with a copious amount of inesthetic but effective epoxy glue.
The entire box was cooled using a PC watercooling pump and CPU heat block. The cooling fluid was refrigerated using a H85DC12 mini water chiller from Hangzhou Purswave Technology, which is a nifty little refrigerator that runs on 12V and features a miniature BLDC compressor whose power can be adjusted using a pot on the controller. The fluid was a roughly 50/50 mix of water and propylene glycol used as antifreeze. One advantage of propylene glycol is its safety - nontoxic and nonflammable - which helps control the consequences of the unavoidable spills and leaks. Instead of the H85DC12 which costs about $200, it should be possible to put something together based on a kitchen freezer (~$0 off the curbside or classifieds). Cooling the entire box introduced the new problem of condensation on the beam window, which was fixed by gluing a resistor onto it with thermal epoxy to locally heat it.
Overall, this prototype was rather kludgy, but was able to take the aluminum block to -40C with a heat load of a few hundred mW and keep it stable to under a mK. This was used to observe the beat note between a Rohm RLD63NPC5 laser diode, which was emitting 641nm at room temperatue, and a HP 5501B stabilized HeNe laser at 633nm using a bladeRF software defined radio. Red diodes seem to be particularly amenable to temperature tuning over a large wavelength range, as shown in the Tech Ingredients video "Cryogenic Laser". This confirmed an instantaneous linewidth of a few dozen MHz in each mode, which decreased with the diode current (including after the diode became strongly multi-mode), and also showed a somewhat concerning unstability of the optical frequency of few hundred MHz per second. This could be due to the poor thermal connection between the diode and the aluminum block with the thermistor, and is one of the points that the next prototype addressed.
The setup is essentially the free-space version of the lanthanum spectroscopy experiment, with the key differences listed below.
The laser diode was a SHARP GH04850B2G which had been wavelength-selected into the 493nm-495nm range. As noted above, cooling 505nm diodes only goes so far and after a point the wavelength remains the same and only the forward voltage increases; nominally 487nm-488nm diodes such as the GH04850B2G unfortunately exhibit the same behavior. The GH04850B2G model was selected by buying a few random single-mode low-cost 488nm diodes from Aliexpress vendors and checking their detailed spectrum for single-frequency regions, as explained above. Wavelength selection was performed by CombinLasers; being able to operate diodes over a wide temperature range makes the wavelength selection easier and lowers costs. You may be able to give the wavelength selection shop a wider range than 493-495nm, it is basically a trade-off between wavelength selection costs and your chances of success. If you are lucky, a random Aliexpress diode may already be at a usable wavelength.
For the hollow cathode lamp, the cheapest barium lamp from Aliexpress (with neon buffer gas) turned out to be adequate. Spectral lines at 493nm and 650nm were clearly present in its output when measuring it with a grating spectrometer.
The beam was slightly misaligned to enter the lamp at an angle (which is acceptable since the plasma region inside the cathode is quite wide), for three reasons.
First, at visible wavelengths, we do not have the luxury of a good optical isolator. Telecom butterfly packages come with a nice micro-isolator built-in, but at visible wavelengths you typically get a clunker whose space it occupies on your optical breadboard is only matched by the size of the hole it burned in your wallet. And, without a good isolator, any retro-reflection into the diode will destabilize it, so the angle drives the reflections from the hollow cathode lamp (in particular its window) away from the diode.
Second, the single-frequency optical power available from the GH04850B2G is only a few milliwatts (as measured with a UNI-T UT385), so we recycle the light from the Fresnel reflection on the lamp's window to provide the reference for the lock-in amplifier (TEFD4300 free-space silicon photodiode, but otherwise identical to the lanthanum setup) instead of throwing it away.
Third, this may help avoid etalon effects from the lamp's window.
The beam was chopped by sending it through the blades of a miniature 12VDC fan nominally designed for cooling small electronics. By adjusting the angle that the fan makes with the beam, obtaining a duty cycle very close to 50% is possible.
The diode's temperature was controlled and monitored by a Sinara Thermostat. The coarse wavelength of the laser was continuously monitored with a grating spectrometer. And the optogalvanic signal near 493nm could be observed - every single time the laser was scanned over a certain temperature! (about 3.6C on this particular diode, but it of course varies)
Part of the reason the signal was so weak is because the effective laser power became rather low due to an unfortunate failure of the silica gel condensation prevention technique, which had worked very well until then but Murphy's law kicked in. The next iteration of the laser diode mount includes a humidity sensor :)
The first step for driving Ca+ at 397nm was to select a diode that would reasonably run single-frequency. But the previous technique of driving a single spatial mode diode just above threshold does not work. A dozen samples of various 405nm laser diodes all exhibited the same nasty behavior: multimode right above threshold with lots of mode-hopping and mode partition noise. So the prospects of getting spectroscopy-grade 397nm laser light were looking pretty grim.
A suprising answer is to use a high-power, broad-area, multi spatial mode diode and drive it at low current, at a small fraction of its nominal power. In this regime, the diode builds up what a little-known paper from 2009 calls a "supermode" and outputs a single optical frequency. I could observe this unexpected property using a scrapped many-watts 450nm laser head (model unknown) from a laser cutter. It would effortlessly put out over a hundred milliwatts of single-frequency light when driven just above threshold.
In supermode regime, the beam quality is poor and the linewidth is currently unknown. Still, I thought it was worth a try.
The previous experiment was repeated using a neon-filled calcium lamp and a SHARP GH04V01A2GC (nominally 1W) diode operated in a single-frequency region where it would put out around 35mW. If that matters, the diode was running without its window - it had been damaged during installation in the collimator, but luckily the semiconductor chip was intact so I cleared the debris from the optical path and continued.
There was a second curve ball as the GH04V01A2GC emitted a lot of ASE which was efficiently absorbed by several wide neon lines in the vincinity, resulting in a strong optogalvanic signal whenever the laser was turned on and which drowned out any signal from the calcium lines I had hoped to observe. This theory was quickly confirmed by replacing the calcium lamp with the argon-filled lanthanum lamp I had used previously, where the argon's ASE absorption was much more reasonable. And using this setup, I could observe lanthanum lines near 399-400nm, see for example the result of a scan near 399.6nm with the La II hyperfine structure visible.
After an argon-filled calcium lamp arrived, the experiment was attempted again. The ASE absorption problem was not present, but I could not thermally detune the GH04V01A2GC far enough to hit the calcium line. Cooling it to even lower temperatures could be a solution - unlike the 488nm diodes it does not seem that a hard limit is quickly reached with those GaN diodes. But instead of building somewhat unwieldy TEC stacks and refrigerators and/or dealing with dry ice or liquid nitrogen, I decided to go for wavelength selection and also replaced the diode with a SHARP GH04W10A2GC (350mW nominal) which emits less ASE and where the process-dependent wavelength distribution happens to be broader so diodes with wavelengths at 25C much shorter than the nominal 406nm are readily available.
Due to the diode damaged during mounting and suboptimal thermal design of the first prototype, I decided to improve the laser mount. The main part is the machined metal block shown below:
The laser diode is inserted into the center hole and soldered using indium-tin eutectic alloy. This solder melts at only 118C which is survivable by the diode and especially TECs - most TECs contain bismuth-tin solder which melts at 138C and causes overheated TECs to separate in two without warning, which can be followed by several hours of very tedious work to clean up their leftovers. The solder easily provides a stable mechanical attachment with excellent thermal conductivity. The block is made of copper for thermal conductivity and because copper is easily wetted by the solder. Working with indium-tin solder is very similar to the common lead-tin solder, except that InSn solder wires typically don't contain flux but it turns out that flux pens made for electronics soldering works just fine. Since the temperature is so low, the tip of the flux pen can be used for shaping the melted solder without burning the pen tip, and making a nice joint all around the diode for the best possible thermal conductivity. InSn is also much more expensive than PbSn or other common solders, but you only need a tiny amount per laser.
At the right of the diode is a hole for mounting a TDK B57861S0103F040 thermistor (10K@25C, B=3988K). The hole is filled with MG Chemicals 8329TFS thermal epoxy, then the thermistor is inserted, and the epoxy cured by heating the copper block to 80C-100C. The general idea of this design is to maximize the thermal conductivity between the diode and the thermistor.
At the left of the diode is a hole for venting the space between the diode and the collimation lens. Care must be taken so that this hole is left open when soldering the diode. This is for preventing humidity from being trapped there and condensing when the assembly is cooled down below the dew point.
For mounting the collimator, on the other side of the copper block is a tapped M9x0.5 hole, which is the de-facto standard that manufacturers of cost-effective collimation lenses seem to have converged on.
Four holes on the sides of the block can be used to insert electrical wires for the diode and the thermistor and glue them with thermal epoxy, so that the wires are temperature-stabilized and temperature fluctuations from outside that are transferred directly to the diode or thermistor by conduction through the wires are reduced.
The part was designed with the open source parametric CAD program SolveSpace (download design file) and machined by PCBWay.
This block was glued to a 20x40mm TEC1-06308 (common low-cost model available from multiple manufacturers) using thermal epoxy according to this Tech Ingredients video: the copper surface was sanded on top of a mirror using sandpaper of successive 240, 400 and 800 grit, and then a fine layer of epoxy was applied using a credit card. The other side of the TEC was glued to another machined copper block (download design file) with a thermistor inside, itself being cooled by a 40x40mm TEC1-12709. That second TEC was glued to a 40x40mm CPU watercooling copper heat exchanger. Note that when sanding those, one must be careful not to sand the part where the silicone hoses attach (which could result in a leak) - if they are decentered as can happen on those cheap parts, sand the opposite side.
This assembly was placed into a 125x80x58mm IP68 aluminum stomp box. The bottom of the heat exchanger was fitted with thermal insulation in the form of a 40x40mm PTFE chair pad floor protector. It was then glued to the bottom of the stomp box using Araldite 2014-2. Adhesion of Araldite to PTFE is hit or miss; if the assembly separates, apply Loctite 770 primer followed by Loctite 406 cyanoacrylate, which reliably results in a strong bond. Wear gloves when handling the Loctite 406 - unlike the superglue from the hardware store, it is very difficult to detach it from skin.
The box was equipped with a beam window (a microscope cover glass has convenient dimensions and is extremely cheap and easily available, though a bit on the thin side), a GX20 15-pin circular aviation connector for the various electrical signals (made airtight using Araldite 2014-2), and quick connect fluid couplings with shut-off valves (FL6PM-C, ML6HC-C) for the liquid cooling.
To avoid back-reflections into the laser diode, the laser assembly and/or the window must be assembled at an angle. Pick whichever you find less aesthetically unpleasant :) I could easily get ECDL flashes on the GH04W10A2GC by deliberately aligning the microscope glass window (with a steady hand) to send back the Fresnel reflection, so there is definitely a lot of sensitivity there (Play with this at low current to avoid damaging the diode!)
As before, the air inside the box was dried using silica gel. A common and low-cost DHT22/AM2302 humidity and temperature sensor was placed inside to check for humidity leaking in and avoid any condensation incidents. Simply running the example from the DHT Arduino library by Rob Tillaart provides convenient readout over a USB connection. With a couple bags of fresh silica gel, the humidity would drop below a few percent within ten minutes of closing the box.
See photos of the assembly: 1 2 3
By using this new mount with a wavelength-selected GH04W10A2GC laser diode, the 397nm wavelength and below were easily reached. With the same optogalvanic spectroscopy setup as before (see photo), it was possible to observe two lines: 1 2. I am pretty convinced that those correspond to the desired Ca II transition from the ground state, and another Ar II transition nearby. However, due to this laser's propensity for mode hopping and hysteresis and general bad temperament, and for the lack of a sufficiently precise wavemeter, I won't venture as far as saying which signal corresponds to which line :) I do own an old Burleigh WA-20 which could in theory resolve these lines, but it is an a utter piece of junk that only works on a good day and irritates me every time it needs realignment, which is often. So, maybe this will be for a day when I feel particularly patient, or after I figure out a better way to measure wavelengths.
Overall, this laser isn't too stable, and exhibits plenty of hysteresis which makes the observation of the spectroscopy lines rather difficult and unreliable. It sounds like careful systematic analysis of the temperature/current plane and its effect on the optical spectrum, as done on W's Laser Page, is strictly required to make this one reliable. This in addition to the poor beam quality, uncertain linewidth, interfering nearby argon line, and barely visible wavelength currently make it a second choice after the 493nm barium laser for a low-cost laser cooling demonstration.
These experiments demonstrate that optogalvanic spectroscopy of transitions commonly used for ion laser cooling is possible using vibration-insensitive lasers made with easily obtained and inexpensive components. Additionally, these lasers are single-frequency; among the other advantages explained above, this property makes FM spectroscopy and offset locks fairly straightforward, so that one laser can be automatically locked to the hollow cathode lamp while another similar laser is offset-locked to the first and performs laser cooling in an ion trap.
Further work is needed to ascertain the linewidth of the 493nm and 397nm lasers, though the results at 633nm are encouraging and show a linewidth commensurate with the natural linewidths of calcium and barium ion cooling transitions. It will also be interesting to measure the optical frequency stability that can be obtained with the new laser diode mount without closed-loop feedback. These questions may be the subjects of future articles.