While the last section looked at MOSFETs, this final section will cover some of the results from my experiments on lighting technologies. The first section will look at the humble NE-2 neon indicator bulb and trying to build a nice I-V curve from actual data. The next section will explore the possibility of running conventional mains-powered light globes from high voltage DC – something that is not ordinarily done. I will also check out the transient performance of the 2450 – a set of tests which previously turned up some anomalous results from previous tests (which I think I might understand why) and conclude with an updated opinion that summarises what has changed through the process of re-evaluating the Keithley 2450 SMU after its return from repair.


NE-2 Neon Bulb Tests

In my original review of KickStart 2 I-V Characterizer, I did a single “Voltage Sweep” I-V curve of a humble small NE-2 neon bulb rated for about 0.5mA at 65V. That was a bit troublesome to do with the SMU originally due to the overheating issue which took multiple attempts to get a good result, so I decided to repeat the experiment.

This time, I changed the parameter to do a “dual sweep” so that it goes both forwards and backwards. This illustrates the range of strike voltages over multiple sweeps. After it strikes, the voltage rapidly falls to the running voltage of about 65V. When the voltage is on the reducing part of the sweep, the I-V relationship is much more consistent down to about 0.3mA when the neon suddenly extinguishes.


But as you can see, this doesn’t have a very good resolution around the strike point because of the time taken to take samples but also because this is being swept in “Voltage Sweep” mode. As I’ve learned from the previous MOSFET tests, the “Current Sweep” mode is very powerful in these circumstances.

In this test, instead of limiting the current at 1mA, I decided to push up to 10mA as I figured the neon could probably handle it in the short term. It seems the bulb heated somewhat, which affected the shape of the curve over successive sweeps. But in this case, we can see the low-current arc instability, the “knee” in the negative resistance portion of the curve which acts as a “stable” operating point for the neon and can contribute to its use in various memory/logic/voltage-regulation circuits. Interestingly, I discovered when the arc is unstable, the result is that you can hear the neon “hiss” and “squeal”. The lowest stable current after striking is about 0.2mA, but on the downward sweep, below 0.3mA it seems to extinguish. Running at about 0.4-0.5mA is probably a wise choice.


I’ve been wondering what happens when an NE-2 bulb is tortured in the short term. While mine are rated for about 0.5mA, I wondered what might happen if I push up to 100mA through it – will it fail right away? What might the voltage look like? I decided to sacrifice two NE-2 bulbs in the name of science.

The result is a rather messy chart. I started with a sweep to 50mA since I wasn’t “game” to start at 100mA, but the bulb survived. The first sweep looked normal up to about 10mA but then the curve started to reverse and waver. I suspect this is because the bulb has heated to the point that the temperature change is affecting how the bulb is operating. A second sweep at 50mA while the bulb is warm saw the voltage increase steeply at 10mA, reaching about 130V instead of the ordinary 75V.


After that, I decided to try running 100mA and that produced rather interesting results. The curves are all over the place, some of them showed oscillations and very low voltages at higher currents which suggests that potentially the neon was operating in an arc mode. But since the SMU is carefully limiting the current, it meant that the bulb didn’t just spontaneously explode. But after a few tests, sometimes the voltage increased (probably since I didn’t allow time to cool on some runs).


After the torture test, one neon showed noticeable darkening on the inside glass, potentially a sign of electrode sputtering as it did quite a bit of time in the “unstable” point when striking. Both showed some discolouration on the leads, because the bulb was likely to be very hot operating at about 7W in such a small globe. In fact, the test clips were slightly melted.

After all of this testing, it seems one bulb may be open to atmosphere, failing at the glass-metal seal as the bulb no longer strikes with even 210V applied over it, suffering permanent damage. But the other bulb (aside from slight darkening and increased strike/run voltage) seems to still operate. It seems NE-2 bulbs are hardier than I expected!


Mains 230V AC Globes on DC?

As a bit of a solar-power person myself, I’ve always wondered about lighting and light-globes in general as they are a very common load on such systems. Many mains-rated globes state 220-240V AC on their packaging, but could they potentially run on DC? In some rare circumstances where you might have a high voltage battery bank and appropriate switches, might it be possible to skimp on having an inverter altogether? I know for a fact that there are some T8 fluorescent electronic ballasts which claim to be compatible on DC through to 400Hz aircraft power, some of which have particular restrictions on DC voltage to ensure a reliable strike. These might be useful for use in high-voltage UPS DC battery banks say in telecoms or other industries. But these are not the norm, generally speaking.


That only meant one thing - gathering up all the spare light globes using a variety of technologies (filament, CFL and LED) that I could find in the house for a quick test. This is probably not so much of a test of the SMU but more taking advantage of the fact that the SMU is capable of running as a high-voltage DC power supply (although with limited 100mA current output). After all, I don’t have any other single power supply (even with all rails in series) capable of putting out 210V which is probably why I haven’t tried this before.

The test involved an I-V Curve in the Voltage Sweep mode, but this time, the sweep was in the negative direction, starting at 210V (using the “extra” margin of the SMU’s 200V range) down to about 50V. This is because many globes require a higher voltage for initial striking but can run at lower voltages – but if supplied a low voltage and swept upwards, may not start-up (especially CFLs). For two of the globes, the 100mA current limit was reached, resulting in a “corner” of the power curve being “cut off”.


It’s no surprise that filament globes (dashed) operated just fine from DC. Being dead-simple as a piece of temperature-resistant resistive wire in an evacuated envelope, the output is very sensitive to the voltage with two identically rated filament globes showing about 1W of difference in output at 210V. The LED globes (solid) had a consistent power for the most part, accommodating voltage variations well up to the point of the regulation failing, because of their constant current nature of operation. This makes them quite suitable for “rough” DC voltage powering, at least for those with proper regulators. Those using capacitive droppers would not work as they rely on the AC waveform to function correctly, but none of my tested globes used such a configuration. The exception was the Lightway globe which seems to never reach its rated power – as a dimmable globe, it may have been confused by the lack of phase-angle information due to the DC supply and the reduced voltage probably didn’t help. For that globe, below about 105V DC, it started to flash as it may not have had enough trickle power supply to keep the current driver IC operational. The CFL globes (dotted) were all not able to reach their rated power output with the reduced voltage and were sensitive to the voltage. Initial power was slightly higher for most as well due to the initial filament warm-up current being drawn, while at lower voltages, the arc could extinguish and the globe could lock in a pre-heat mode of operation which could damage the filament if exposed to this condition for prolonged periods.


I only discovered two or three CFLs in my collection that don’t operate from DC. Two of them draw almost no current from the SMU when given 210V, while the other pre-heats its filaments continuously but cannot strike the tube which may also lead to destruction of the tube in time. I suspect the latter may rely on the mains AC for timing or the 210V DC wasn’t enough to strike the tube as its voltage drop may have increased due to age, but the results seems encouraging.


This is perhaps not surprising, as most ballasts may be taking in AC, rectifying it to DC and then “chopping” that up to drive the tube or LEDs so sending DC to the globe doesn’t really change much from the perspective of the driver. I suppose if you have some “high voltage” DC (e.g. a UPS battery bank) and the right switch-gear, maybe some commodity mains AC globes can operate without the need for an inverter but I wouldn’t rely on it. Still an interesting experiment (at least, to me) which satisfies one of my long-held curiosities.


Transient Performance

In the original Instrument Performance in-depth blog, I found some unusual results when testing the transient performance of the Keithley 2450 SMU. While the B&K Precision 8600 DC Electronic Load can generate very fast current rise/fall time waveforms, the Keithley 2450 SMU typically fared better than all power supplies I had tested to date, but the slow waveforms were showing some voltage drift which seemed unusual to me.


This time around, I decided to improve on my instrumentation set-up to try and eliminate some influences. Four-wire connections were previously made using stacked banana plugs which I later realised may have influenced the read voltage as the stacked banana may be conducting current and internal electric fields may have caused some voltage offsets. This time, I used individual banana plug to clip wires of equal length to the terminals of the B&K Model 8600 DC Electronic Load. The R&S RTM3004 oscilloscope was originally connected using an x10 passive probe, but since we are measuring low voltages, I decided to “forego” the extra bandwidth and use an x1 passive probe to allow for lower noise. I’ve also used the front-end 20MHz bandwidth limiter and high-resolution mode to reduce noise further on the traces.

The fastest transient with an 11.6us rise time showed a 977mV peak-to-peak deviation in the voltage while the SMU is outputting 10V. The regulation appears to sort it out within 325us. This is a significant deviation, but given the challenge, most power supplies would see the output collapse completely to zero for a while rather than just a relatively short 10% dip. The downward load transient causes a deviation of 1.048V but the deviation is taken care of in an even shorter time of about 120us. Given the datasheet states a “settling” time of <200us typical, these results seem rather consistent with the claims and the result is not much different to what I had seen in the previous testing.

Looking at a slower 1ms rise/fall time transition, noisy due to the load itself. The result seems to show the output voltage deviating about 30mV average during the ramp on a 10V output. The baseline can be seen to shift about 4mV. This is more than the voltage source accuracy of 0.015% + 2.4mV and datasheet claimed voltage source noise of 2mV. However, I would say this measurement is perhaps inconclusive as we’re measuring very small values on an oscilloscope and my experience is that values close to 10mV or below are difficult to correctly resolve.


Perhaps if I had an expensive power-rail probe, I would be able to more accurately measure the voltage offsets and noise.

Testing the very-slow transitions, the peak-to-peak voltage was about 13mV. The deviation does follow the current waveform but the amplitude is smaller, so perhaps the error has to do with my test set-up. The periodic higher-frequency noise is visible in the high-resolution mode, with an amplitude of about 10mV peak-to-peak, but whether this is from the SMU or inducted into the test set-up is not known. I could not push into a finer vertical resolution scale, as the individual samples contain “spike” noise that is clipping and the “high-resolution” sample averaging is “hiding” that noise.


On the whole, the results seem better than previously tested, but that may be due to improved care in the test setup. The overall results again confirm that the SMU has a tight and fast regulation loop compared to other power supplies I’ve tested, but with the probes I have access to cannot achieve high accuracy at the millivolt level. Ideally, an expensive power-rail probe would be used to attain clearer and more conclusive results.


Conclusion: My Updated Opinion

As this is the final update blog post for the Keithley 2450 SMU after the RMA process, I thought I’d end this section with a conclusion that looks at what has changed since my original review.


On the whole, I’d have to say that the majority of the conclusions I came to with the original review still hold true post-repair on this re-evaluation, thus the original review still remains intact. The issue on the 200V range has now been resolved which means that a number of tests are now possible which would have previously caused overheating messages and channel shutdowns. Some of the issues are still awaiting the next firmware release to be finally resolved. The resolution of this issue has also meant that I no longer see “Overflow V” results in KickStart 2 which is a good thing.


If anything, where my follow-up tests did change my opinion would perhaps be that of the Keithley KickStart 2 software and the usefulness of TSP scripting. Now that I’ve had an opportunity to spend more time using KickStart 2, while its intention is to be an easy no-coding point-of-entry into testing with Tektronix/Keithley instruments, it has quite a bit more flexibility and capability than I originally understood from my relatively limited exploration of the software initially. With careful use of the configuration options, it is possible to run tests which may not seem possible (e.g. trending over time using I-V Characterizer) and it is possible to configure plot axes and traces in different ways to make good sense of the data. The project management features in terms of having project descriptions, possibility of orchestrating multiple instruments, having per-run descriptions, storing settings and data in .ksp/.hdf5 files with batch export functionality makes it useful enough that I often used KickStart 2 just to “get the answer” rather than try to develop something bespoke.


But when I did need to develop something, the power of TSP scripting with the TestScriptBuilder environment makes it easy to code, debug and develop something that can take full advantage of the SMU’s speed and precision. It was considerably easier than I expected to write something as complex as a battery-simulator proof-of-concept and a capacitor testing script. While I didn’t iron out all of the bugs, it did produce something that was very usable and even had the capability to log files to USB without taking more than a weekend to assemble at the most.


A minority of the original gripes may indeed be a case of mea culpa as this is the first SMU I’ve had the fortune of using for more than a couple of hours. Unlike ordinary power supplies, the sourcing and measurement are an integral part of the solution, so auto-ranging doesn’t quite work as you might expect (e.g. auto-ranging changing ranges causes the output to become undriven for a short period) and certain settings need particular care (e.g. output off setting). My expectations, based on my experience with one- and two-quadrant power supplies, were perhaps not entirely applicable to SMUs – for example, metering behaviour with the output switched off which may oscillate or show unusual values especially as the SMU is capable of extremely low ranges which can be dramatically affected due to pick-up of stray voltages. Likewise, while an SMU is capable of operating as an (expensive) power supply, it isn’t all that efficient at doing so with a limited output power. Instead, it is much more suited to precision voltage/current sourcing/measurement for characterisation tasks where power demands are usually quite low or for short periods and accurate limiting is a must. This can definitely be seen in my current calibration testing where when sourcing/sinking about 400-600mA, the unit gets quite warm and the fan is pushing full speed.


On the whole, after the repair, the Keithley 2450 SMU now operates as it should and is one of very few choices of SMU on the market today. As the original review stated, there is little reason to look elsewhere as the 2450 offers equal or more at the same price as its competitors, but with a lot of flexibility. Unfortunately, when it comes to using I-V Tracer on the 2450, I’d have to say that the resulting application is still a little lacking and the license is perhaps better used deployed to a higher end AC-capable SMU such as the 2461. The SMU has been a lot of fun to work with and has given me quite a few insights when it comes to running I-V curves on various components and devices – I am so thankful to have one in my possession at last.


This concludes my re-evaluation of the Keithley 2450 SMU as part of the RoadTest program. Thanks for reading my posts – I hope the findings and opinions have been useful and interesting to you all. Last, but not least, thanks to Tektronix/Keithley and their representatives who worked extremely hard to make sure the hardware issue was addressed properly within their permit and offered a full license to Keithley KickStart 2 in goodwill.


Oh … and Merry Christmas everyone. Here’s hoping for a better 2021!