For most instruments, this chapter is perhaps the most exciting one as I delve into the instrument’s performance capabilities across a number of parameters. However, this time around for some metrics especially, I have been thwarted by two considerations – the first is the instrument’s exemplary capabilities which sometimes well exceed the capabilities of my test equipment suite, and secondly, the issues experienced with the instrument itself which may make these results atypical of regular 2450 units.


*PLEASE READ* - This Review is with Caveats!

Unfortunately, the SMU I received appears to have a hardware fault causing problems with its 200V range and also produces some occasional unexpected readings even at lower voltages. I have been in contact with representatives from both the Singapore and US branches of Tektronix since 22nd May 2020 reporting my findings and attempting to work through the issues towards a resolution. However, in part due to COVID-19 disruptions, it did not appear that a full resolution of the issues would be forthcoming prior to the deadline for review delivery.


While reading this review, please keep in mind that the unit I received does have hardware faults which may mean that the instrument performance results are not necessarily typical of properly functioning Keithley 2450 SMUs. Testing actually began with this section first, which was when I first detected the potential hardware issue which progressively disrupted other parts of the review as well. It made me question whether it was worth running these tests in the first place, as these results may not be representative in the case that my 2450 is faulty.


Instrument Problems

Because this chapter concerns itself with instrument performance, I thought I’d start with the symptoms of the issues I have seen with the particular 2450 I received, so that readers can better understand the remainder of the results in the context of the faults observed.

The first signs of issues occurred during the voltage/current programming and measurement testing when I was testing the 200V range. As tests were run unattended for the most part, I noticed upon waking one morning that the output had tripped off overnight when sourcing a voltage of about 65-70V. Reviewing the system log on the 2450 revealed a heatsink 1 overtemperature event which appeared to be spurious, as the unit was sourcing 200V into a ~1-10Mohm DMM input which is an infinitesimal load. I reported this to Keithley/Tektronix immediately, however, rerunning the test did not cause the fault to recur, so I initially dismissed this fault as possibly a transient, one-off event.


Later testing with KickStart 2’s I-V Characterizer proved to be much more troublesome, as any load on the output during a sweep past 60-85V would consistently result in odd behaviour. This included transient overheat events, also affecting heatsink 2.

Whenever this would happen, the output would be tripped off to protect the SMU from damage. Suspecting a possible software issue and with confirmation that a firmware downgrade would not damage the instrument or I-V Tracer app license, I reverted to the version of firmware provided at the point the unit was shipped. The output still tripped off, but it did not raise an overheat warning.

This particular fault manifests itself in different ways but once hitting the “magic” threshold voltage which varies from run to run, the unit buzzes, hisses and relays inside click. Errors are sometimes logged, overheating and/or source limits. The display also shows unexpected values – e.g. a sourcing of +86.2V is requested with a limit of 100mA, but the unit is reading -170.857V which is in entirely the wrong quadrant and much greater than the test limit! In the other shot, Overflow A is reported for current while the voltage is at the upper OVP limit of +211V while +75.8V is requested.


In fact, during testing, I tried loading the output with the B&K Precision Model 8600 DC Electronic Load which is capable of 0-120V only. The unit tripped reading a similarly high negative voltage – but thankfully, it seems this may not have actually been output to the front panel jacks or the unit managed to survive it. Another inconsistency also shows with the output light on the front panel remaining on in some cases, while the output appears to be isolated.


Oddly enough, if I attempt the same conditions under a no-load condition, it is almost impossible to get an instance of the output tripping out. It only seems to happen under load – this can be an infinitesimal load such as the gate of a MOSFET, or a larger load like a DC-electronic load set to 10mA.


Since provoking this behaviour a number of times for troubleshooting reasons, the unit began to exhibit new unexpected behaviour. In auto-ranging mode, it starts to continuously report Overflow values for amps or volts which I did not see before and did not expect (as it should change ranges on the first instance of occurring it). I was also somewhat unfamiliar with the instrument’s regulation behaviour in auto-ranging mode and was surprised when the set source limits were not always respected and the “limit” annunciator appeared – the actual limit also depends on the range which auto-ranging chooses and a change in range results in the output being lost momentarily (at least on my unit), resulting in perplexing results when attempting dynamic load tests. This is resolved by choosing an appropriate manual range.


Keithley/Tektronix have been informed about these findings, along with my questions about its behaviour and currently suspect a hardware issue which requires instrument service (affecting my unit as well as at least one other RoadTester) and are in the process of working out a timeline for a resolution which has been affected due to COVID-19 restrictions and staff being on leave at times. Regardless, I still felt it important to deliver a review due to the community’s anticipation, in the spirit of full transparency and to remain compliant with the terms and conditions of the RoadTest program.


Voltage/Current Programming and Measurement Accuracy

Assessing the accuracy of the Keithley 2450 SMU is a tough challenge. As a precision instrument, its datasheet capabilities often match or surpass even some bench-top single-purpose instruments. As I am a hobbyist, not an accredited testing facility, the amount of gear I have is necessarily limited and, in many cases, outside the calibration window.


For this testing, I have enlisted my Keithley 2110 5.5-digit DMM, last calibrated from the factory in 2013. Because it is so far outside the calibration window, its accuracy cannot be guaranteed, but I have some confidence that it is still generally “within” its error margin based on cross-checks with other instruments, at least to the 4.5-digit level. The 2450 SMU poses additional challenges because of its 6.5-digit capability, so to give the 2110 the best chance, I’ve chosen to warm it up for at least three hours prior to each run, use 10PLC (the most accurate setting) and average of 16 readings (to remove noise but without increasing the test time too far). Not all ranges could be tested on the 2450 because the 2110’s measurement capabilities would not be sufficient especially in the low current ranges.


Combined voltage programming and readback accuracy is based on assuming the 2450 SMU’s output is exactly the value which is requested and instead recording the difference between the requested value and what the 2450’s metering is reporting the output to be. The error margins plotted are based on the measurement error specifications alone, which is more stringent than the combined effect of both programming and measurement errors, but this was the best way to resolve the fact that I did not have any way to know with a high level of confidence and precision what the true output value of the 2450 is.


As a result, the results from this section are less about judging the 2450 SMU’s accuracy but more just to confirm the 2450’s in the right ball-park and a “check” that all is well with my 2110. It is also an opportunity to run multi-week-long command sequences to evaluate the stability of the instrument’s remote-control interfaces.


Voltage Programming Accuracy

In the 20mV range, the 2110’s readings are relatively flat and well within the 2450’s error margins, indicating that the 2450 is well within specifications. It seems that the offset is about 20uV which could easily be the result of thermal EMFs.


In the 200mV range, the 2110’s readings are still relatively flat, with the error bands encompassing the zero-line indicating a high likelihood that the 2450 has virtually no output error. The range of error of the 2110 is contained entirely within the 2450’s range of error, indicating the 2450 is very likely to be better than specifications imply.

Moving up to the 2V range, the 2110’s different ranges start to show an impact and errors at the edge of the ranges begins to show up. The zero-error line is still encompassed within the range of error of the 2110, with the range of error of the 2110’s readings almost entirely contained within the 2450’s specifications providing high confidence that both instruments agree and the 2450 is well within specifications.


At the 20V range, a similar result is shown with the edge of ranges on the 2110 showing some discontinuity, but the zero line is within the range of error on the 2110 and the complete range of error of the 2110 is contained in the error margins of the 2450, indicating the 2450 is easily within specifications.

On the 200V range, testing required two attempts due to spurious overheat warnings. The final result shows the zero-error line being encompassed by the margin of error of the 2110, indicating that the 2450 is likely to be well within specifications. Above 120V, the 2110 is less accurate than the 2450’s sourcing capabilities, so it is hard to draw firm conclusions in this range, but the data suggests it is within specifications.


Combined Voltage Programming & Readback Accuracy

Reading the value from the 2450 and comparing that with the requested voltages shows that the 2450’s own measurements of its outputs confirm that it is well within the specifications throughout all ranges.

In the 20mV range, the error is well within specifications and the error seems to be confined to under 2uV.

The same trend is seen in the 200mV range, with the error confined to about 10uV.

In the 2V range, the measurement claims the error to be within about 40uV.

Using the 20V range, the error is around 400uV.

Finally, at 200V, the error is about 4mV. The results are excellent when compared with the datasheet specifications and clearly illustrate the instrument’s precision capabilities.


Current Programming Accuracy

Testing the current programming accuracy really shows just how difficult it is to accurately measure current on a DMM. In the 1mA range, the output error of the 2450 is much smaller than the measurement error of the 2110, however, interestingly all of the differences lie within the output error range of the 2450 which suggests both instruments agree and are still performing better than their specifications imply.


The 10mA range shows a similar result, with a bit of a slope in the error line. This is not unexpected as the 2110 seems to have a bit of a gain-error which can be seen in the voltage readings which show discontinuities at range changes, but this is still within the specifications for the DMM. No problems here!

Moving up to the 100mA range, a very similar result is shown although there is a slight “curve” to the result, which may be due to heating of the shunt in the 2110. The absolute difference lies well within the 2450’s error margin, although the 2110’s error margin is larger especially at higher currents.


Testing the 1A range shows that the absolute error is within the 2450’s error margin with some gain error visible, but the 2110’s error margin is still similar if not larger. The absolute error between the two instruments is still only about 0.5-1mA at a current of 1A which is very impressive – I suspect most of this is from the lower-end out-of-calibration 2110. The precision of this instrument is above any other sourcing instrument I’ve seen.


Combined Current Programming & Readback Accuracy

Similarly, with the voltage combined tests, the current combined tests show that the 2450’s own measurements suggest the output to be well within the claimed error margins.

In the <=1mA range, the current error is within about 20nA of the programmed value. This illustrates the power of the SMU to be a “true” current source – most regular power supplies have current limiters but the error in current can be quite significant (into a few mA).

By the 10mA range, the error is within about 200nA.

At 100mA range, the error is mostly within 2uA.

On the 1A range, the error increases noticeably, now reaching about 50uA, but still tiny compared to what conventional power supplies usually achieve.


Warm-Up Drift

Warm-up drift was assessed by using a Keithley 2110 5.5-digit DMM that had been warmed up for six hours in a room with stable temperature, measuring in the 100mV DC range at 10PLC. The Keithley 2450 SMU was powered up from cold, having rested at least six hours and the output set to 50mV with the recorded voltage plotted over a period of an hour.

The voltage produced initially starts slightly above the warmed-up value, falling slightly below before flattening out. While one hour of warm-up appears to be recommended, stable outputs seem to be achieved in close to 25-minutes. There are a few spikes in the result, which seem likely due to mains power quality issues with ripple signalling coming on and off, which may have affected the DMM or the SMU in some way. The final voltage was about 5.5uV above the set value which is well within the margin of error of the 2110 used for measurements, with a total deviation of no more than about 10uV total excluding spikes, representing excellent stability overall.


Unexpected Power-Down

The behaviour of the 2450’s outputs when the power is removed using the hardware power button was investigated using the Rohde & Schwarz RTM3004 measuring the channels under no load except for the 10:1 oscilloscope probe of 10Mohms.

At an output of 1V, unexpected power down causes a loss of regulation initially, resulting in a slow linear voltage decline, followed by complete loss of output. However, if sourcing negative voltage, the unexpected power removal causes a similar loss of regulation but then the voltage briefly becomes positive at a similar magnitude, before output is completely lost. This suggests that unexpected power-down is not that gracefully handled in the case of sourcing negative voltage.

The same experiment was repeated for 200V to understand the supply’s behaviour at the extremes of its operating envelope. The positive case is practically identical to the 1V case. In the negative case, the voltage rises to about 70V positive before declining. It seems prudent to switch the output off first before using the mains power switch to avoid potential damage to devices when sourcing negative voltage.


Dynamic Load Regulation

Dynamic load regulation was tested using a B&K Precision Model 8600 DC Electronic Load to generate variable slew-rate transients, with the 2450 running in four-wire sense mode. The 2450 was set to source 10V at up to 1A. Voltages were measured from the sense lines using piggy-back banana plugs into the RTM3004 oscilloscope in AC-coupled mode. All measurements are made using a 20MHz bandwidth filter and the high-resolution mode. Load steps were made between 100mA and 900mA of load on a periodic basis.


The results are somewhat inconclusive, due to the noise in the set-up which may have been due to oscillatory interactions between equipment, external interference and maybe even issues with the 2450.

Under a constant current load of 500mA on an output of 10V, the current monitor line from the B&K 8600 exhibits some oscillations while the AC-coupled output exhibits noise typical of the 10:1 probe, external noise pick-up and front-end effects combined. This illustrates why it’s not possible to measure ripple and noise using such a set-up. Enabling high resolution mode allows us to “cut through” some of the noise by averaging samples. This reduces the effective sample rate and seems to show a set of “blips” every 500ms. This seems to be coming from the 2450 SMU itself, as I was not able to reproduce this using another power supply, but could just be a measurement artifact.

Challenging the 2450 SMU with the highest rate transitions with rise times in the 10-15us region, a deviation of close to 800mV was recorded. On load application, recovery occurred in about 250us and on load removal, recovery occurred within 75us. The deviation may seem high, but transition rates this fast regularly bring ordinary power supplies to their knees, collapsing the output to zero for a considerable period, so this performance well surpasses anything I’ve witnessed to date.

Using a slower transition time of around 1ms, the 2450 SMU seems to track the current changes including oscillations from the 8600, but there seems to be an offset in voltage at the new load. My expectation was that the voltage at both ends of the load ramp should be virtually identical because the measurements are made on the four-wire sense legs, but there seems to be close to 10-20mV of difference.

At a much more leisurely transition time of about 130ms, the voltage offsets are quite a bit clearer and seem to follow the inverse of the load curve suggesting they are likely somehow ohmic. I did not expect this when operating in four-wire mode, but I have noted some other regulation quirks and the “noise” blips from the 2450 SMU, so perhaps this is not typical performance.


CV/CC Transitions

To test the 2450’s constant voltage to constant current transition behaviour, I set the 2450 to source voltage and measure current, selected an output of 200V but with a limit of 100mA. The output was connected the output to a 4.7-ohm 10W wire-wound resistor and the RTM3004 oscilloscope to observe the output on power-up.

The expected voltage is about 470mV, so the output request of 200V is far above this. The SMU handled this excellently, as the output ramped up rapidly in the space of 15us, with a minor level of overshoot to 600mV, which was extremely brief, lasting under 5us. This was relatively difficult to provoke – I tried it at 20V and was unable to see any overshoot despite the rapid rise time. This shows the 2450 SMU’s speed and accuracy which far surpasses general-purpose and even performance-level power supplies.


Output Generation Speed & Rate

Testing of the speed of output generation was done using the RTM3004 measuring the output of the 2450 while the output was commanded on and off using the front panel. The time taken for the output to rise to the programmed voltage and to fall back to zero is measured.

Two different speeds of output generation were observed for generating 1V at 0.1A. This may be due to selecting auto-range, resulting in the range current limit restricting the rate of voltage rise, or due to a presence of an internal “sleep” mode (which I’ve met on other instruments in the past). Regardless, in the slow case, the output took around 312ms to rise, while in the fast case, this was around 40us.

Turning off the output under this setting took about 312ms to achieve with a smooth linear ramp down to zero.

Likewise, repeating this experiment for a 200V 0.1A output revealed two different speeds for the output rise. The slow case actually sees a non-monotonic increase in voltage, first reaching about 110V, declining a few volts before sharply rising to the target voltage in 155ms. Perhaps this is related to the fault with the instrument as noted in the caveats section. In the fast case, the output rises in a near linear ramp in a matter of about 1ms.

Turning off the output at 200V sees the output fall in a near-linear fashion over a period of about 4.51ms.


The rate at which outputs can be generated was assessed using a simple TSP script that had a while loop calling functions repeatedly to set the SMU’s output voltage between two levels. The resulting “square” wave output was measured with the RTM3004 to determine its output rate and slew-rate capabilities.

Toggling between 1V and 5V produced a square wave of about 440.309Hz (2.271112ms). This means that each output dwelled for 1.135556ms. This is fast, especially given that the wave seems to be relatively “square” still, indicating we are not being too limited by slew rate capabilities. While some other instruments may advertise arbitrary waveform generation up to 1kHz (1ms), they may be limited by slew rate which means that the 2450 is probably just as good, if not better, in this regard.


Trying my luck, I pushed the voltage to -200V to 200V, trying to sweep the full range of the SMU’s capabilities. The limited slew rate did not allow the unit to reach the target voltages, while the frequency of the wave fell to 340.344Hz (2.938204ms), implying a speed penalty for multi-quadrant operations. This is perhaps no great surprise, so while it seems to do its best to try and achieve a 1.469102ms dwell time, its output is slew-rate limited.

Remaining within the same quadrant restores similar timings to the above example, however, the effect of the slew rate limitations makes the wave look more like a sawtooth. Operating solely in the positive voltage sourcing or negative voltage sourcing regime seems to produce nearly identical results.


Digital I/O & Interlock

The digital I/O was tested with a square wave input signal from the RTM3004’s signal generator, driving a TriggerFlow program that used the digital I/O to turn the output on. Time was measured from when the input signal was sent into the digital I/O port and when the output begins to be generated.

The digital input was very snappy, with the output rising to the full output value within 400us of receiving the input.


Testing was also performed with the interlock, this time with a manually generated assertion and de-assertion which would trip off the output. Time between the toggling of the interlock state and the output being turned off was measured.

Asserting the interlock, the output was turned off within 1.5ms. De-asserting the interlock resulted in the output turning off within 3.55ms. This is not quite as fast as the digital I/O, but both are very fast compared to the other instruments.


Power Consumption and Efficiency

The power consumption and efficiency were tested using a Tektronix PA1000 Power Analyser to measured the consumed power at an input of 230V, 50Hz AC from a pure-sine-wave inverter source. The output was sunk into a Rohde & Schwarz NGM202 Two-Quadrant Power Supply at a range of power levels.

When powered up with the output switched off, the power consumption is around 25W. Once providing power, it seems the power consumption increases linearly at a rate of 2.76W for each Watt of power delivered, reaching about 80W when sourcing 20W. This results in a relatively low peak efficiency of 25%. This is, unfortunately, one of the downsides of having a high precision sourcing instrument – the precision comes at the cost of efficiency.


Acoustic Noise

My subjective assessment of the acoustic noise is that the power supply is not too loud. The fan is not usually audible unless under extreme stress conditions of delivering or sinking maximum current. Under that condition, the noise is definitely noticeable and has a minor tonal element making it sound like a louder laptop with its cooler fan running at full blast with a characteristic whine. The spread of the vents does ensure that good ventilation regardless of positioning, with the fan placed deep inside the unit rather than at a side panel possibly to reduce the amount of noise emanation.



The determination of the instrument’s performance is made difficult by the fact that the unit I received is suspected of having a hardware fault, thus is not a typical unit. Notwithstanding this, the 2450’s specifications are generally higher than the instruments I have at my disposal, thus firm conclusions on some parameters can be difficult to draw. Regardless, I believe it was still a worthwhile exercise, even if it was time consuming, so that I could gain some understanding of how SMU performance compares with ordinary power supplies and to have the opportunity to test the unit’s remote-control interfaces over long command sequences spanning a number of weeks.


In terms of instrument stability, in general, I had no issues with remote control using SCPI over VXI-11 LAN for the purposes of the instrument testing, in fact, the Keithley Model 2110 5.5-digit DMM connected via USB used to perform comparison measurements faulted twice in the same timeframe. The testing, however, was interrupted by spurious overheat errors which tripped the output off, which is related to the suspected hardware fault. The instrument generally remained quiet under testing, but when stressed, the fan nestled within the unit becomes audible like a louder laptop with a tonal element manifesting as a noticeable whine. It is not subjectively loud, but is noticeable.


The output programming accuracy in voltage was verified with the 2110 which was able to clearly indicate that the output error was within datasheet ranges despite the 2110’s calibration being 6 years out of date. This is a good indication that both instruments are doing well. When the 2450 is used to judge itself in a combined programming and readback accuracy test, the errors were well within specifications with errors ranging from about 2uV to 4mV depending on range. This level of precision is generally unattainable with even performance level power supplies.


The output programming accuracy in current was more difficult to verify, due to the difficulty of accurate current measurements in general. The margin of error of the 2110 generally was at a similar magnitude or greater than the 2450’s datasheet specifications, thus not allowing any firm conclusions to be drawn, however the absolute difference lines did remain close to zero and generally within the 2450’s margin of error which again illustrates the likelihood that both instruments agree to within their margin of error. Using the 2450 to gauge its output, the errors were well within error margins, achieving errors within 20nA to 50uA depending on the range, however, low current ranges below 1mA could not be adequately tested due to limitations in my test equipment. This illustrates the additional capability of the SMU to be a “true” current source, providing fine, accurate current sourcing capabilities which is generally not achievable with ordinary power supplies.


Testing of the warm-up drift showed a very small amount of drift – on an output of 50mV, the output drifted about 10uV in total, settling in to its final value in close to 25-minutes.


Unexpected power-down behaviour of the supply for positive voltages is good, showing a loss of regulation initially resulting in a slow voltage decline followed by the loss of all output. However, for negative voltages, the initial voltage decline is followed by a short burst of positive voltage before loss of output, meaning that unexpected power down when operating in the negative voltage regime could result in unexpected output, and it is prudent to turn off the outputs of the 2450 before using the mains power switch.


Dynamic load regulation results showed a very rapid regulation response, where slower transitions were tracked by the power supply, including the 1ms and 130ms examples, although there seemed to be some voltage offset at the two load levels of 100mA and 900mA which seems to be ohmic in some regard but should not have appeared especially with the use of 4-wire mode. This may be related to the various instrument quirks. When challenged with the most demanding transitions with times in the 10-15us region while outputting 10V, the supply showed a deviation of about 800mV with recovery in 75-250us. These rapid transitions easily bring most power supplies to their knees, collapsing their outputs to zero with significant recovery times, so in light of that, the SMU handled this case well. There was some output noise and oscillation visible in high-resolution mode every 500ms which seemed specific to the SMU which I did not observe with other instruments – it could be related to the combination of test equipment or the test set-up, so this result is a little inconclusive.


As expected, the power supply has great constant voltage to constant current transition behaviour. It was difficult to provoke misbehaviour, with no overshoot when asked to source 20V/100mA into a 4.7-ohm resistor (470mV). Increasing the voltage to 200V managed to cause a brief overshoot to 600mV for under 5us, despite the rapid application of the output, rising within 15us.


Output from the SMU has been observed to either be “quick”, rising up to the output value within 40us to 1ms, or more leisurely, ramping up over approximately 155-312ms. Turning off the output results in a smooth near-linear ramp of a period of about 4.51-312ms, dependent on voltage. The output toggle rate appears to be fastest when working within a quadrant, achieving around 1.1ms dwell times for each point. Crossing quadrants, the dwell time was closer to 1.47ms. Attempts to do a -200V to 200V step wave showed slew rate limitations which prevented the SMU from reaching the programmed output voltage. However, 0-200V step waves were possible, with rounded edges looking more akin to a sawtooth wave. While the speed based on a simple TSP loop is not as fast as the 1ms dwell time which some performance power supplies advertise, the faster slew rate of the SMU actually makes its output far more accurate even if its maximum output rate is a bit slower.


The digital I/O performance was very snappy, with the input of digital trigger to output generation propagation time of just 400us. The interlock signal was responded to within 1.5 to 3.6ms.


The downside of the SMU is that its power consumption and overall efficiency are lower than an ordinary power supply. At 230V, 50Hz supply, the 2450 idled at around 25W with each Watt of output costing about 2.76W, reaching about 80W when sourcing 20W. The peak efficiency measured was 25%. While efficiency may not be the primary concern of an SMU user, this is perhaps something that is not obvious and is to be borne in mind in case you intend to be using many of them for automated testing in a rack due to power and thermal concerns.



This blog is part of the Keithley 2450 SMU with I-V Tracer RoadTest.