[RFC] Driver: current driver for laser diodes

[jordens]

We may be looking at designing a diode laser current driver in the future. Comments and secondments welcome.

General notes

• Low-noise current controller for (low-power) laser diodes.
• Form factor TBD. May end up being a Stabilizer mezzanine or a separate board that interfaces with Stabilizer or a separate EEM
• Either two versions for 200-250mA and 1.5-2A range or variants or one design that fits both
• Different from the coil current driver in that it has smaller range, drives diode-like loads, doesn't need that much stability, but needs modulation features.
• Linkage to Thermostat: ethernet, Stabilizer: analog, and Zapper: analog
• Specifications and performance goals similar to the open designs:
  • Libbrecht-Hall
  • Libbrecht-Hall-Durfree
  • Libbrecht-Hall-Seck
  • JQI to be consolidated, based on Libbrecht-Hall-Durfree
• Compare to proprietary things:
  • Vescent
  • ILX Lightwave
  • Toptica
  • Koheron: uses LT3045, LT3261, OPA567, LT6655
  • MOGLabs
  • SISYPH SMC11
  • AOSense controller module

Variant A

• 250 mA
• 2V compliance
• single quadrant
• assumptions/requirements w.r.t cathode grounded, anode grounded TBD
• 0.5 µA resolution, digitally controlled (~100 kHz resolution, 200MHz/mA tuning), may be relaxed significantly to ~16 bit, depending on laser characterization
• 10 kHz small signal bandwidth
• RMS noise 10Hz-10MHz: 0.3 µA
• RMS noise 10Hz-10kHz: 0.05 µA
• Stability (<10Hz) TBD
• Linearity/absolute accuracy: 2% TBC
• tempco 5 µA/K (1 MHz/K, 20 ppm/K) TBC
• high demands on laser diode safety guarantees: no turn on/off transients, no chance of ESD/ground potential difference getting through to the diode, benign behavior and recovery from short circuits and open connections
• could be some coarse+fine additive or multiplicative hybrid solution to achieve resolution TBD
• has a modulation/error signal input that is both sampled by an ADC (at 100 kHz) and a high frequency bypass (10 kHz to 100 MHz) that goes onto a bias tee (together with the low frequency current output from the controller) on the output to the laser. Or alternative approaches TBD.

Variant B (otherwise same as A)

• 1.5A
• 3 V compliance
• bandwidth and resolution much less critical
• noise requirement at a similar relative level as A

[dtcallcock]

Out of interest will you make this compatible with Toptica laser heads (either pin compatible or via an appropriate adaptor cable)?

a high frequency bypass (10 kHz to 100 MHz) that goes onto a bias tee on the output. Or alternative approaches TBD

The Toptica lasers do the fast modulation and bias tee on a little PCB right in the laser head. This lets them provide modulation ports as coax connectors on the laser head so that you don't need the main cable to be that high performance. Seems a reasonable approach.

[dtcallcock]

Form factor TBD. May end up being a Stabilizer mezzanine or a separate board that interfaces with Stabilizer or a separate EEM

If we make this as a mezzanine and we also make the piezo driver mezzanine, should enough signals be passed through so they can be stacked? This would presumably require finding suitable surface mount headers so they can be placed on both sides of the board without interfering.

[jordens]

The target here is a bit more complicated than the toptica lasers but I don't see an issue with making a wiring harness for that assuming they didn't deviate from the old LMU style too much.
Whether you bias-tee in the driver or at the head is a choice. What is important is that this is in general not just a tee but a four-port element (wide band error signal in from e.g. a phase lock split into lowpass out to the controller, highpass out to the laser plus lowpass path from the controller to the laser). That makes this element as cumbersome to have in the head as in the driver.
Stacking boards more than two high is something I would avoid at great cost. The mechanics, the electrical issues and the pain to debug it easily are so annoying. I would prefer to have the piezo driver a separate two-port amplifier device, especially since Stabilizer already has a suitable output on the front panel. But we won't have a piezo anyway.

[jbqubit]

@restelli may have comments based on his experience with the JQI roll.

[gkasprow]

A long time ago I designed some laser driver. It had an interesting feature - I used depleted MOSFET to ensure the laser diode is shorted all the time during on/off transients, even with power off. The MOSFET was disabled when safe operation of the diode was ensured. We can use the same mechanism here.

[gkasprow]

@jordens can you sketch how the coupling of the diode with modulation input and high-frequency bypass input could look like? Does it have to support grounded cathode and grounded anode? If so, maybe the design should be bipolar...

[jordens]

Generally speaking it's only about feeding a modulation signal (some "error" signal) to the diode where the low frequency part (<100 kHz) is filtered by the diode current controller. I.e. the high frequencies go through unmodified (we might talk about adding just a VGA/VVA and a relay/switch in that path if possible) while the low frequencies go to the controller (ADC), which then has a low frequency output (DAC) that is combined with the high frequency part and feeds the diode. This analog bypass allows fast feedback and fast modulation of the diode while the current controller can both sense and steer out the low frequency error signal (potentially onto some other channels like diode temperature or VHBG) and at the same time have extremely good low frequency control of the current.

       --[LP]--<ADC-DAC>--[LP]--
      /                         \
MOD --                           -- LASER DIODE
      \                         /
       --[HP]--[VVA/VGA]--[SW]--

The modulation input is only HF coupled to the diode. It's grounding arrangement doesn't matter.
The diodes we are looking at are floating. I.e. the current controller can be any type. There might still be unequal capacitive impedance to ground so the feedpoint of the HF modulation component should probably still be selectable (anode or cathode).
Alternatively -- and one could argue that is better -- the MOD-LP/HP splitting could be pushed elsewhere. Then there would be a HF input at the head and a LF input at the driver.
For a general application we'd need the DC part to be configurable as AG or CG but in my experience that comes with quite some complexity. Similar complexity to making it bipolar.

[sbourdeauducq]

The diodes we are looking at are floating.

On what lasers?

[jordens]

Micro-integrated ECDL/MOPAs

[sbourdeauducq]

It would be more interesting if the current driver would also work with less expensive and niche options. Do more common designs also have floating diodes? The newfocus tunable lasers do, but I don't know about others.

[jordens]

If the "less expensive and niche options" contribute to the development then they should step forward and do so. They need to convince users that they are better. If you are referring hobbyists, then this is probably not interesting for them. It's simply not what's funded.

[gkasprow]

If we want bandwidth of 100MHz then the signal needs to be injected to the diode very close to it. It's hard to make a broadband matching circuit for such non-linear device. The diode resistance depends strongly on the bias current.

[jordens]

Sure. That RF signal only needs a reasonable wideband match from 0.1-10 MHz. Between 10 and 100 MHz there would be "actual" narrowband synthetic modulation signals that don't need a particular match as they can be tuned in power and frequency. Just that above 10 MHz the filter should not attenuate intentionally.

[dtcallcock]

With a simple bias-T you should be able to get ~GHz of bandwidth. In the Toptica lasers this is a separate input to a well-matched DC-20MHz input. I like this approach as you can use one input for your servo and the other for your modulation.

[sbourdeauducq]

I know hobbyists are boring and broke, but there are proper academics with proper grants who still make their own lasers (or use commercial laser heads) and could buy the device. It would expand its market.

[jordens]

It's certainly not nearly a flat response up to GHz. As mentioned, the stuff above 10 MHz (and arguable above 100 kHz) is not really within the scope of the current servo here. The laser head people need to account for that.

This is just that: a proper academic project with a proper grant that makes its own laser. Anybody (also without academic job and without money) is welcome to speak up and contribute good arguments, analysis, insight and work.

[jordens]

To clarify the AG/CG question: The driver needs to be of one such type. Either driver type is compatible with floating diodes and the respective grounded type of diode. For the other diode type, we'd need a bipolar driver or change the design for the other polarity. That added complexity and design cost/risk/time (for bipolar or second design) would need to be offset by the users of the "less common" diode connection scheme. TA/PAs are typically common anode, while the MO diodes are more likely common cathode. In our case both are floating but the anode has larger ground capacitance but that will be checked later.

[gkasprow]

Take into account EMI. If you drive a diode assembly with the anode connected with housing and grounded cathode, your device will act as an antenna. So, for RF we need to make sure that housing is always grounded for RF signals.

[jordens]

That's what I said in https://github.com/sinara-hw/meta/issues/37#issuecomment-509635714 as well and it applies to some extent to floating diodes as well.
But it's not that complex to add a mechanical switch to feed the HF into either cathode or anode. It is however complicated to design the driver for bipolar or switchable polarity.

[gkasprow]

True, the bipolar driver is difficult. It's much easier to design positive and negative drivers separately.

[jordens]

If this ends up being a single design that can do either grounding scheme, then I would put both drivers (or their final stages) into the same device (the positive on the anode and the negative on the cathode) and deactivate/bypass/ground the unused driver side. That way it's just pseudo-bipolar which actually matches the way the diode itself operates (the diode is not bipolar just the grounding scheme is).
And the HF modulation part can be T-ed onto the hot end of the diode at the head. The LF modulation can be then sensed at the head and its sign be treated in software.

[gkasprow]

We can actually make a module with separate outputs for both types of lasers.

[gkasprow]

Is Libbrecht-Hall-Durfree design sufficient for this application?
There is a room for improvement, one can replace the output MOSFET with FET + BJT output stage to further limit the noise. I'm not sure if we can really benefit from Stabilizer design. It's easier to place the control DAC directly at the current source opamp input. Stabilizer generates GND referenced signal which needs further conversion since we want supply-rail referenced voltage. I'd rather use 16bit DAC as in original design. It's easier than using the output of the Stabilizer. Do we need complex AFE as in Stabilizer?

[restelli]

Neil Pisenti and I a few years ago designed a low noise laser driver for the JQI strontium lab. The novelty of the architecture is that is designed to be ground referenced and to use extremely low power yet maintaining a relatively high control bandwidth in closed loop configuration. Noise characterization and overall design are available in our RSI paper: https://arxiv.org/abs/1609.03607

The project is on Github: https://github.com/JQIamo/hv-piezo-driver

Several of these current controller are in use in the JQI strontium lab, and I will be happy to show them in action to @JQubit.

[restelli]

The main limitation of the JQI architecture is that for high amplitude sweeps it has a small slew rate.

[gkasprow]

@restelli it seems you posted to wrong issue. You probably meant Zapper issue

[dnadlinger]

one can replace the output MOSFET with FET + BJT output stage

I was about to mention this, having toyed around with a few current source topology ideas recently (no finished designs). Cascoding the FET with a fast BJT should improve the output resistance at higher frequencies, and hence reduce the wideband noise. Polarity-wise, this appears considerably easier when sinking current (fast PNPs are disappearing rapidly, while you can e.g. get very fast and linear Si:Ge NPNs).

Might also be worth considering using a BJT after the op-amp instead of the FET for the lower 1/f noise (with bias current cancellation if necessary).

[jordens]

@gkasprow The design is sufficient if it meets the spec ;)
I'd also go with a BJT. 16 bit current resolution won't be sufficient here. 0.5µA/250mA is 2e-6.

[gkasprow]

If we need more than 16 bits, the Stabilizer is not sufficient anyway.
@jordens what about ADC front end spec?

[gkasprow]

@jordens
0.5µA/250mA is 500k which means we need 19bit DAC. There is no much choice and we would have to use expensive AD5791 DAC

[jordens]

If it was just about current capacity, wouldn't they just use an IGBT?

[gkasprow]

IGBTs are intended for switching applications and are not noise-optimized. It was probably easier to take 50 low noise fast BJTs than one large slow one.

[jordens]

About the approach to switching the polarity: I really like it if the DAC operating from some artificial, shifted ground doesn't cause problems. Just needs some tweaking on the output relays, protection, terminals, and making sure the unused driver side doesn't wreak havoc into the reference rail buffer.

[gkasprow]

Such DAC operation was in the original Libbrecht-Hall design. It minimizes the number of opamps.
But indeed, we may end up with 2 independent circuits due to complications from a switched polarity approach.
What about making a single polarity source and then, use inductors on both wires and AC coupling + switch to determine if the modulation goes to anode or cathode? The circuit could be simplified if we assume that the laser is tied to +5V in one of the configurations. Then we get rid of the switch at all.

[jordens]

Sure. The HF modulation can just be routed by a switch and AC coupling to the hot end. But I don't see how you drive the LF/DC on the hot end of a common anode/common cathode diode in any other way than with two different driver architectures.

[gkasprow]

I was talking about 2 scenarios:

1. we use one driver and passive LC circuit to inject LF and HF components. But this probably won't work because diode assembly would pickup the noise.
2. we use proposed architecture but common anode assembly would be connected to 5V instead of GND.

[jordens]

• (2) Will work fine with our laser and all common cathode lasers. For a grounded anode where the laser case is also ground, that would put the laser case, cable shielding etc to 5V which isn't going to fly.
• (1) I don't get the "one driver". How do you drive the DC current?

[gkasprow]

In (1) the DC current will always flow in one direction which has implications that HiZ output would be on the diode case wieh common anode. One can easiy damage it with ESD. Let's skip this idea because it's silly.
It also means that it's not worth to think about 5V on laser case (2)

[jordens]

Yes. The ground of Driver and the ground of the laser head are common. It's either (a) build a driver for floating and grounded cathode and exclude common anode or (b) build two drivers or a single driver that can be configured to source or sink (with positive voltage or negative voltage respectively).
I don't think the polarity switch implementation above is bad. If Libbrecht-Hall works for CG, then it can be made to work AG as well (modulo PNP-vs-NPN differences).

[gkasprow]

This is exactly what I did. I merged Libbrecht-Hall CG and AG versions to minimise the component count

[dnadlinger]

@jordens: Heh, fair enough; I guess I've spent entirely too much time thinking about building low-noise photodetectors recently… To be fair, though, were it not for a couple of problems (1/f noise is a gamble, woefully underspec'd at DC, only two manufacturers still packaging discrete ones), GaAs FETs really be the perfect device for low-noise transimpedance frontends – the guy has a point.

@gkasprow: I realised that the sources in your sketch were just a stand-in rather than something you wanted to pass off as a high-performance/ultra-low-noise/greatest-thing-since-sliced-bread design, hence the word "placeholder". ;)

Regarding topologies: I've first looked at that Scandurra et al. photovoltaic gate drive current source paper a few years ago now, and am still waiting to get a chance to make use of it. Its main advantage (apart from being able to use a N-channel device) seems to be that you can easily retain good output behaviour while being able to trivially reduce the control bandwidth almost arbitrarily far to suppress control voltage noise. The floating source has some µV / rtHz noise, so in their example, they go down all the way to 1 mHz corner frequency! Not quite what we are after here, I think, but I haven't done the numbers on far you could push the design for redimensioning it for 250 mA / 10 kHz. (The optical coupling bandwidth might also become an issue.)

As for other architectures, maybe the most straightforward design is just to use a large sense resistor after all. If we take this to the extreme and allow ourselves to burn 5 W in the sense resistor (using something like the Vishay VPR series), we could go up to 80 Ω for 250 mA. At this point, 90 pA / rtHz is more than 7 nV / rtHz, so we can just directly use an op-amp + FET for the whole bandwidth. (~47 Ω off a ~15 V rail is probably more realistic.) With a few hundreds of kHz of modulation bandwidth, one could directly use a single path for narrow-linewidth laser locks. No base current to worry about as well, so getting the required accuracy is easy.

(Modulation sidebands would still be applied directly via capacitive coupling – the equivalent resistance obviously varies between diodes, but something like 10 nF and a 3:1 transformer should be in the right ballpark. Do commercial laser heads match the modulation input better than that? I never dared to look.)

[dnadlinger]

(Ah, 50Ω is actually what is used by the Libbrecht–Hall design used, too – hardly original or very daring, then.)

[gkasprow]

@dnadlinger do you know this idea?

[dnadlinger]

@gkasprow: I looked at the Dudzik paper a while ago. Splitting up the driver is potentially nice – a LT3045, say, gets you most of the way there (2 nV / rtHz, so the local sense resistance can be small), which, among other things, should allow you to use a much smaller and easier to drive FET for the last few µA. As with most compound things, I'd be slightly apprehensive about getting the AC behaviour at higher frequencies right, though – for instance, what does C4 in Grzegorz Dudzik's circuit do to the loop dynamics?

This wouldn't change anything about burning a considerable amount of power in the sense resistor to get down to the required levels of current noise, though, as per the above estimates.

Overall, I don't really have a set idea about the way to go on this yet – too many critera of unknown weight (power dissipation, physical size, modulation requirements, …).

---

(An aside regarding that paper: I don't see how his comment about the AD8605 being "optimised for low source impedance" make sense; it is not central to the claim in any way, though.)

[dnadlinger]

(I also haven't checked how much the output inductor actually helps to suppress high-frequency noise in the published designs given the finite source bandwidth, i.e. lower resistance at higher frequencies. The estimates on the required bare noise level might be pessimistic.)

[gkasprow]

One must make sure that inductor is well shielded and won't pick up the noise.

[jordens]

@dnadlinger As mentioned, the Koheron uses extactly that LT3045.

[jordens]

Funded by QUARTIQ/Creotech/FBH/U Warsaw.

[dnadlinger]

Nice! (Out of curiosity: FBH as in https://www.fbh-berlin.com/?)

By the way, regarding the discussion, one aspect that went unmentioned so far (if I recall correctly) is the modulation depth; at least, it is missing from the draft specifications. Like in the original Libbrecht–Hall design, keeping the main current source at a low bandwidth and then adding on a higher-bandwidth (but small) component will make reaching the goals much less taxing.

[dnadlinger]

Another commercial (?) offer: Sisyph SMC11

[jordens]

Yes. that FBH is making the modules.

The modulation bypass+sampling is discussed a couple times already (regarding cross-over, (non-)matching, protection). Or do you mean something different?

Thanks for the additional link. Added to the top post.

[dnadlinger]

Or do you mean something different?

Just that building a current source that can be modulated from 0 mA to 250 mA at 10 kHz with > 18 bit precision and low noise is considerably harder than if the modulation depth on an otherwise static current is only, say, 0.25 mA max.

[jordens]

IMO the current tends to split into more than just two components: (a) a coarse offset with low resolution (1e-3) and very low bandwidth (kHz) (b) a small steering current of about 1e-2 of (a) with high resolution and medium (~100 kHz) bandwidth to do feedback (including integral) and compensate instabilities and INL/DNL of (a), (c) a narrow bandwidth high (but pretty much unknown) frequency analog modulation from MHz to GHz with low demands on matching stability and matching
quality (d) a wide bandwidth (10k-10M) matched bypass for fast feed-forward and low-f modulation (the DC-10kHz part of which is sampled and handled by (b)).
The sum of all terms needs to be low noise according to the specs above. Merging any of those groups for easy implementation is fine. Splitting them because implementing their combination is challenging is also fine.

[dnadlinger]

Agreed. I just meant to point out that there wasn't a draft spec on the amplitude for (b) (and, possibly by extension, (d)).

What "small" means pretty obvious for physicist users, but not having that spelt out might make the corners of the design space less clear from EE perspective. Anyway, I'll sit back and let you all get on with this now. ;)

[jordens]

Added the AOSense module to the top post. Thanks Mariusz for pointing it out.

[dtcallcock]

Is there any reason to think Variant B wouldn't also make a nice current supply for the few-Gauss field coils found on most ion traps (or at least a good starting point for one - a factor of 2 more current might be nice)? Is it funded or was that just the A Variant?

[jordens]

You'd have to deal with the inductive load. Both variants are funded.
But for shim coils it looks like overkill: around 10 A-turns you can dimension the coil to be in the range of a simple single-opamp voltage-to-current converter running of Zotino/Fastino since there are no grounding restrictions. This would also be bipolar while Driver will not necessarily be.

[fimek]

Right now there are following problems regarding laser driver:

• combining the expected noise level with reasonable power consumption,
• laser driver precision (meeting the 16-bit DAC precision and resolution),
• bipolar transistors instead of unipolar transistors (due to their more promising noise parameter),
• DAC noise density within 100 kHz bandwidth (seems to be the most challenging - the DAC cannot be BW limited for noise reduction).

It should be decided whether it is sensible to test just the unidirectional current source, or the current source sould be designed as bidirectional (compatible with both common anode and common cathode lasers).

[fimek]

My strategy briefly explained:

1. DAC: non buffered DAC, where the only output noise is the one generated by the R-2R ladder. Best achievable parameters: 7...9 nV/sqrt(Hz). Proposed ICs: DACx1001, AD5791 (seems similar in performance).
2. DAC output voltage swing: maximum available: 0...10V. If the reference voltage for the current source is too high - can be divided leading to DAC output noise reduction. 0...10V will match the current source optimum driving voltage.....
3. ....which depends on the output transistor emiter resistance. 40 Ohms seems OK, together with ref. voltage of up to 10 V result in 250 mA of maximum laser current.
4. Current source has to be built around the opamp with En <= 3nV/sqrt(Hz), low input current noise (together with DAC ladder resistance would degrade noise performance) <= 0.5 pA/sqrt(Hz), low bias current and appropriate input voltage offset. Fortunately - such opamps exist: AD8675.
5. By using the additional JFET we can get rid of output transistor base current and maintain DC stability.
6. Power supply: +15V /- 5V (can be +15V/-15 V either).
7. It's a bit tricky to generate a current source with grounded anode/cathode, because the reference voltage for the current source must be referred to Vdd or Vss. I'd rather avoid differential amplifiers (due to remarkable additional noise contribution). I' would use LTC1043 (or similar) instead (the BW is sufficient).