Help Understanding ESD Test Failures in a Battery Powered Device

Thread Starter


Joined Dec 1, 2017
In the interest of detailed explanations, this is going to be a very long post so I’m sorry in advance. Specific questions are located at the end.

Short Version:
I have a battery powered device that is failing ESD testing. There are TVS diodes built into the system but I’m still seeing voltage spikes in the hundreds of volts on power rails. As far as I can tell no damage is occurring to the hardware itself however, there are erratic operational behaviors of the device.

Long Version:
I’m having some ESD problems with a device I’m designing at work and, after a week of researching and fiddling, am stumped. I’m the only engineer at my job and am quite green, having graduated December 2016 (however I’m not new, engineering was a career change for me). This is a device I’ve designed from the ground up, is the first device I’ve designed on my own, and certainly the first that I’ve had to design for EMC. It’s been a case of not knowing what I don’t know, though I’m slowly working my way into the “knowing what I don’t know” territory. I’ll do my best to give a complete picture but can’t simply post the whole design since this is a work project. (I have permission to make this post)

I am designing a medical device that is a multi-channel, constant current, electrical stimulator powered by four NiMH batteries which contains four isolated 5VDC->120VDC flyback converters in an ABS enclosure. It does not connect to any other equipment, only the patient with “stimulation” and “return” electrodes, one of each per channel. Each discrete stimulation cable attaches to the device with a connector that has metal housing. Previously I had the connector housings floating but I have now tied them to their respective channel grounds which has helped some. On that note, my understanding of a device like this is that there is no such thing as ground, only the return path back to the batteries. For ease of typing, I’ll be using the term “ground” even though it’s not technically correct. Channel 1 of the device shares a ground with the rest of the system, but channels 2, 3, and 4 are isolated from each other and have their own “ground.” (Would this be considered a virtual ground?) These isolated grounds are connected to the system ground via a 4.7nF capacitor which is required for emissions per Analog Devices.

The device is being designed to the latest IEC 60601-1-2:2014 standard with ESD testing based on IEC 61000-4-2. This requires up to 8kV contact discharge and 15kV air discharge. When I went in for pre-compliance it passed radiated emissions, most of radiated immunity, and conducted immunity with flying colors. The radiated immunity failure was one very specific problem within the proximity field test, failing at only 870MHz in the horizontal plane. It failed ESD quite badly due to not operating correctly but has been able to withstand the discharges without physical damage to the system. I have rented an ESD gun and 61000-4-2 compliant test table to help sort through these issues. Tests have been performed with resistive loads on the stimulation cables.

I have designed in the LTC4367 to handle undervoltage situations with the intent of protecting the NiMH batteries from depleting to the point of damage where they will not be able to take a charge again. It is also used for reverse polarity protection. I have a SMAJ9.0CA bidirectional TVS diode installed between the positive and negative (“ground”) battery terminals, less than 2mm away from the through-hole connection. Ultimately the issue seems to be that there is no actual clamping occurring, causing huge voltage spikes to reach the LTC4367, violating the valid input voltage window and triggering it to enter its protection/shutdown mode. I have validated this theory by watching the battery voltage, output voltage, and LTC4367 fault output. The fault line drops at the same time as the output from the FET’s, remains low for ~32ms, then comes back online. This follows the operational description of the chip exactly, so it is operating as it should, but causing problems in my design. (See FIGURE 1) These measurements were taken with four scope channels attached and the scope lead ground attached to the device ground (See FIGURE 2). I realize this changes the operation of the system because there is now a path to earth, but the observed behaviors are still present.

When the system is in its standby but turned on, the onboard capacitance seems to be able to hold the microcontroller above its 1.62V minimum operating voltage, which allows it to hold the system power supplies on. The LCD backlight cuts out, but that’s to be expected as its LED driver has a 3V minimum voltage. When the device is stimulating there is a high-power draw (up to 900mA) and the onboard capacitance isn’t nearly high enough to keep the microcontroller operating for 32ms, so the whole system shuts down. (See FIGURE 3) - Note that VBAT and VIN are poorly named in this context. VIN comes from the batteries and VBAT goes to the rest of the system. It was semantical oversight that has since been corrected.

I’ve tried bypassing the MOSFET that the LTC4367 drives and the device handles the ESD much better, though there are still some intermittent issues. When I do a contact discharge on the stimulator cable connector I still see massive 100V+ oscillations on the power rails (See FIGURE 4). The amplitude of these oscillations is relative to the ESD discharge voltage (2/4/8kV). Using an active differential probe across the 10kΩ load resistor I also observe odd spikes in the output current (See FIGURE 5, FIGURE 6, FIGURE 7), though I am not sure I believe what I am seeing because of the nature of my test setup (See FIGURE 8 ). I do use a wire tied to the HCP to discharge the device after each ESD event. I am wondering if those spikes are due to the fields being generated around the probe. Regardless of that, the stimulation current shut down briefly after the discharge and then returns to normal.

We also just added an override to the safety checks in the firmware, preventing the device from shutting itself down when it sees erratic behavior. After doing this we began to see a new behavior where, when the stimulation current comes back online, it shoots up uncontrollably instead of returning to its prescribed value. I designed in a discrete hardware safety circuit that cuts all power to the stimulation circuits if any of them exceed 5mA, and this circuit is triggered when the current output loses control. (It’s a nice validation that the circuit works!) The ramp up 0 to 5mA occurs in roughly 400us. I am unsure of how to measure the DAC output in this scenario because attaching the scope will provide a path to earth ground and I feel like that will interfere with the measurement?

  • Question 1: Because this device is battery powered it can be viewed as floating because it has no true path to earth ground. Thus, when a discharge event occurs, the entire device is charged up to some voltage. So “ground” could be up at 8kV relative to earth and my 3.3V rail would still be 3.3V above “ground.” This is why the standard requires the device to be discharged with an earth grounded cable after each ESD event. Conceptually, is this understanding correct?
  • Question 2: I do not understand why the TVS diodes aren’t clamping where I would expect them to. I know that I have some kind of misunderstanding of their operation in this application and this is a design error on my part. What is this misunderstanding?
  • Question 3: How should ESD protection for a battery powered device like this be approached? Expanding on question 2, my general approach is clearly flawed; do you have recommendations for a more appropriate approach to ESD protection in a device of this type?
  • Question 4: To continue the theme, what is the interaction of the three isolated channels with the main system power and each other during an ESD event. I realize this is probably worthy of a dissertation by itself, but it is also the area where I am most unsure.
  • Question 5: This might be out of the scope of this post, but I’ll tack it on here anyway. Table 7 of 60601-1-2 specifies that “discharges shall be applied with no connection to an artificial hand and no connection to patient simulation.” These tests are being performed while a simulated load is being stimulated. When no load is present the device will not operate. Do I simply not need to be testing the device while it is operating as intended?

I have also attached very general pictures of the basic system, constant current circuit, and HV flyback.
Any help would be greatly appreciated


Thread Starter


Joined Dec 1, 2017
Further thoughts:
I harp on my tech to use the little spring instead of the long ground lead on the scope probe for power supply measurements but I've been doing exactly that. I'll have to get a little creative getting that measurement with the short spring return but should be able to do it.

Some points of clarification:
1. The enclosure is ABS plastic with no paint or EMI coating
2. The metal connector shells are not electrically connected to the enclosure. I have gone and connected them to the respective channel "grounds" which seems to have helped some, but I need to make some recorded measurements with them connected and removed for a proper comparison.
3. All connectors exhibit the same behavior. There is little to no difference regardless of the connector I discharge on or the channel I do a current measurement on.
4. Both positive and negative discharge polarities behave the same way. The only difference I've observed is seeing the voltage spike starting positive or start negative.

Some additional observations I've made since this original post:
The isolated 3.3V for channels 2/3/4 drops out when a discharge occurs, which would account for the output current going to zero because the DAC/Op-Amp loses power. This 3.3V comes from a RECOM R1SX-3.33.3-R isolated converter to cross the isolation barrier. I have ordered some Murata NXE1S0303MC-R7 converters which are footprint compatible and have a higher isolation voltage. I'm not optimistic that this will make a difference but figured it can't hurt. I'll just have to add a load to meet its minimum load requirements.

I have added 0.1uF capacitor across pins 1(VIN) and 4(GND) of the LTC4367 which slightly modified the oscillation waveform after the initial discharge spikes but that was all. I also added a TVS diode in parallel with this capacitor to minimize the loop inductance/return path from the TVS diode across the battery terminals, again with no luck or appreciable change in my measurements. That being said, I question the oscilloscope observations due to my measurement technique. See the attachment where the top capture is my initial measurement and the bottom capture is the same measurement but with a 0.1uF capacitor across pins 1-4. I have also attached an image of my rework method for adding the capacitor.

Additionally, I have ordered a collection of C0G radial ceramic caps that I can try placing in parallel with the assorted TVS diodes. I realize the leads aren't the best thing to have but with the SMA diode packages I'm using I can't get the MLCC's soldered on in parallel without adding wires. To round it all out I purchased some connectors with a thermoplastic shell instead of metal. I'm not sure if my boss/marketing person will go for a plastic connector instead of the nice metal ones, but it seems like a very good way to remove something from the equation which would allow me to use air discharge instead of contact for the entire device.


Last edited: