Posts

One Ferrite Is Not Enough

This would be a great Bond film title…

“So Blofeld, do you expect me to talk?”

“No Mr. Bond, I expect you to solve this radiated emissions problem!”

* laser noises intensify *

 

I was doing some radiated emissions problem solving on a smart LCD module and found an issue that is not new but I haven’t encountered for a while.

In this case, the solution required two ferrites. One on the I/O cable to the module and one on the flexible cable between controller and LCD screen.

Adding only a single ferrite in some cases INCREASED the emissions rather than reducing them, presumably an effect where the addition of the ferrite changes the resonant frequency of either one leg or the entire antenna to the troublesome frequency at 192MHz.

This reinforces the approach of:

  1. Always add new fixes to existing fixes already implemented. Whilst it might be the fifth change that worked, it might not have worked without the previous four.
  2. Once the last fix is in place and validated as working only then can you try and figure out what combination is actually required to solve the problem

The last step can get very busy, particularly if there are a large number of modifications applied. It might only be worthwhile if some are particularly expensive or difficult for the customer to implement in production. Different fixes for different budgets!

 

Graphical Guide to EMC: Near Field Probing (free eBook)

 

Download “The Graphical Guide to EMC: Near Field Probing” eBook here (40MB)

 

I have a love / hate relationship with textbooks.

They are thick, have lots of words, make me feel clever, and stop my bookshelf from floating away. They often have the one thing that you are looking for.

On the other hand, they have far too many (big!) words, too many equations with no context or explanation. I find it very difficult to sit, read and quickly gain an intuitive understanding.

 

I prefer to communicate with pictures. This is why my presentations are image heavy and text light. I’ve sat through far too many “PowerPoint Karaoke” sessions where the presenter reads the words on the slide.

Also I love the format of cartoons and graphic novels but you rarely see them outside of the fiction sphere. I’ve recently been thinking about what a combination of a graphic novel and a text book would look like.

 

With the recent acquisition of an e-ink tablet with drawing stylus to replace my 74 different notebooks and notepads I started sketching out some ideas for a guide to using near field probes. A subject that I’m often asked about and is complementary to our free Pocket Probe Set that we give away at shows and to customers.

One thing turned into another and once I started drawing I couldn’t stop. You can download the full eBook from the link at the top of this page or by clicking here.

 

 

I’ve released this under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license. This means you can share or adapt this work but you must provide a credit / link back to the original source (here). Any adapted work must be shared with the same licence terms.

 

I’d be interested to hear your feedback on the format and content of this mini eBook – please get in touch and let me know. If there is a positive response then more content may follow.

Thanks and all the best

James

 

 

self interference demo USB3 and 2.4GHz

2.4GHz Intra-System (or Self/Platform) Interference Demonstration

In this blog we are going to take a short look at noise and interference in the 2.4GHz band. Our example victim is a Zigbee controller and the sources are nearby USB3.0 devices and Wi-Fi sources.

 

Background

One of our customers makes these rather useful USB Zigbee Coordinator sticks, frequently used for controlling smart home or IoT devices like light bulbs.

These devices operate at 2.4GHz, a very crowded frequency band with Wi-Fi, Bluetooth and Zigbee all fighting for a narrow, congested slice of spectrum.

One of the common issues faced by users of this band is that of intra-system interference, sometimes referred to as “self” or “platform” interference. This is where components in the same system interfere with each other, primarily due to their proximity.

[Note: The counterpart to intra-system (within the system) in this context would be inter-system interference (between separate systems), which is what the conventional EMC test regime of radiated and conducted emissions and immunity seek to characterise.]

This common problem is something that our customer knows all too well from helping their clients integrate these Zigbee products into the end application.

So, during a recent visit to our lab for some testing on a related product, we spent some time investigating this noise on a typical setup.

 

Demonstration Setup

The setup in the below image is common to many users with a Raspberry Pi Model B and lots of stuff plugged in to the USB ports. In this case, a Zigbee adaptor (black case) and an USB3.0 SSD in close proximity.

These parts, including the spectrum analyser, is part of the customers in-house electronics development laboratory.

 

self interference demo USB3 and 2.4GHz

 

The effects of USB3.0 on the 2.4GHz spectrum are well known. A good example is this 2012 paper from Intel which

For this demo, we used a near field capacitive probe and a 2.4GHz antenna to measure noise in the 2.4GHz to 2.5GHz band local to the Raspberry Pi.

This demonstrated the degradation of the noise floor with various levels of system activity including

  • Measurement of system noise floor
  • Presence of a USB3.0 SSD running a large file transfer using the dd Linux command
  • Activation of the Raspberry Pi internal Wi-Fi

The below image shows three traces under these different conditions.

 

spectrum of 2.4GHz band showing ambient noise, SSD noise and Wi-Fi emission

 

Experiment Conclusions

The conclusions we can draw about the in-band noise are:

  • Noise from the SSD raises the noise floor by approximately 10-20dB (a factor of x10 to x100)
  • The Wi-Fi transmission from the Pi is 40dB above the local noise floor. This will mask any received Zigbee signals from a remote transmitter.

 

In-Band vs Out-of-Band Sensitivity

Well designed radio systems are generally very robust to out-of-band interference i.e. anything outside of the narrow radio band that it is tuned to. For instance, a Zigbee radio system set to channel 20 (2.450GHz) will reject anything below 2.445GHz and above 2.455GHz.

 

Intra System Interference Diagnosis

Advice on diagnosing these issues is mostly outside the scope of this short blog. Differences in systems, components and ambient noise levels makes it impractical to offer guidance for all situations. However, some generic problem solving pointers are presented below.

A systematic approach to isolating the problem is required.

One of the primary rules of problem solving is to change only one thing at once and observe the effects.

In EMC terms, it is possible to change several things at once without realising it. Cable position, the specific port that a device is plugged into, location of nearby equipment and cables, even how firmly a connector is tightened will all make small differences that stack up. (Don’t use anything other than a torque spanner on those SMA connectors though!)

Another key rule is if you think something has made a difference, reverse the change and see if the problem re-occurs. Unless you can achieve consistency then you might be changing something else unintentionally, or the problem is caused by something outside of what you are changing.

Correlating the problem against time can help. Does it happen when something else happens (other devices on, or off, or switching, certain configurations, times of day, etc.) This can give clues.

Lastly, we should be looking for a significant step change in improvement to identify the issue. Phrases like “I think it made a bit of a difference but I’m not sure” indicates that we are dancing around the issue and not getting to the heart of it.

Ultimately, for a detailed understanding, the spectrum analyser is a key tool in gaining a proper grasp of this issue.

 

Solutions

The solutions to the problem are simple yet sometimes difficult – a technical balance needs to be struck.

Use of Ethernet rather than Wi-Fi on the Raspberry Pi.

It is not practicable to synchronise transmission from the Raspberry Pi Wi-Fi with that of the Zigbee stick. The simplest way of ensuring the Wi-Fi does not interrupt the Zigbee transmissions is to disable the Wi-Fi and provide network connectivity via Ethernet instead.

Depending on the installation this might not always be practicable but it certainly is more reliable.

 

Separation of components

Moving the antenna away from the noise source is usually the best way to achieve increased performance.

In this instance, placing the module at the end of a USB cable and away from other electronic items is a good start.

Another option that is not as ideal: a good quality SMA extension cable could be used to extend the antenna away from the problem area. This introduces loss into the RF channel, reducing signal quality.  Measurements made in our lab on a cheap extension cable from RS show a power reduction of 6.5dB at 2.4GHz for a 5m cable. This equates to a ratio of around 0.25 meaning we are broadcasting and receiving a quarter of the power we were before.

Also, it is still possible for the noise to couple onto the nearby module even without the antenna attached meaning the problem does not get entirely resolved.

 

Better quality components

Sourcing a bunch of cheap-as-possible parts from Amazon or eBay is likely to bring problems.

Using devices from big name manufacturers and buying from reputable sources helps. But, even reputable components are designed to a price point and can still cause problems if the other points in this blog are not taken into account.

USB cables can be a big source of the problem. Unshielded back shells (the part between cable screen and connector body) compromise the shielding to the point where their performance at high frequencies is equivalent to an unshielded cable.

The only way to tell if a cable is good quality is to perform an autopsy on the ends and check on the cable shielding

Remember that Pawson’s Law of Cable Quality states that the EMC performance is inversely proportional to the physical appearance. Braided covers, shiny plating, metal connector bodies, transparent mouldings etc are all indications of money spent on the OUTSIDE of the cable. EMC quality comes from the INSIDE and is not visible.

shiny usb cable vs boring usb cable

 

 

Hope this was useful! See you soon.

James

 

 

 

conductive contamination underneath surface mount isolated power supply causing line to earth surge failure(marked up photo)

Surge Test Failure Due to PCB Manufacturing Process

We recently had a piece of customer equipment fail the IEC 61000-4-5 surge test at 2kV line-to-earth. There was a loud crack of an electrical arc forming, the unit stopped responding to communications and was making a hissing/squealing noise.

To give it the appropriate technical term, this was “A Bad Thing”.

Using the thermal camera we quickly found several hot components all on the 3V3 supply line that we supposed had been damaged by the surge. The hissing noise was the DC/DC converter in a cycle of burst mode trying to supply too much current before shutting down.

However these were all secondary side components on the isolated part of the system. How did the surge get across the safety barrier? The designer was using correctly rated parts and the PCB creepage distances were dimensioned correctly.

As part of the fault diagnosis process, we used our hot air solder rework tools to remove one of the isolated power supplies providing a low voltage supply to the AC mains monitoring circuitry. Underneath we found this:

 

conductive contamination underneath surface mount isolated power supply causing line to earth surge failure(marked up photo)

 

The samples had been hand soldered by the customer, unfortunately leaving a large amount of solder paste underneath the power supply.

Whilst this was not a short circuit across the safety barrier it did reduce the creepage distance significantly. When a 2kV surge (1.2/50us, 12 ohms) was applied from AC mains to earthed secondary this pollution was enough to cause an arc to form and into the 3V3 supply pin (centre right).

This voltage was enough to fry several components on the 3V3 line, rendering the board inoperative.

 

Lessons Learned

  • Hand soldering prototypes is OK provided you take great care in the process and cleaning the board afterwards
  • Professionally manufactured boards will generally avoid this issue
  • Apply a line-to-earth safety test on your AC mains powered products to check your samples
  • We are going to start a policy of performing a line-to-earth safety test on all AC mains powered products coming into the lab for testing from now on to try and catch problems like this.

 

 

V2 pocket EMC debug probe PCB - near field probe set - board front

Pocket EMC Debug Probe V2

This is a guide for the assembly and use of the Version 2 “Pocket EMC Debug Probe” from Unit 3 Compliance.

Download our free eBook on Near Field Probing

 

V2 pocket EMC debug probe PCB - near field probe set - board front V2 pocket EMC debug probe PCB - near field probe set - board rear

 

Assembly Guide

Components required:

0805 resistors for R1 = 470R and R2 = 10k, gives 450 ohm parallel combination. This is required for the 10:1 into a 50 ohm input.

0805 capacitor C1 470pF to 10nF, C0G/NPO dielectric, 50V. For the AC coupling of the signal.

Sourcing SMA edge mount female connectors (RS, eBay)

Recommendations for the probing pin (socket strip or a bit of wire)

Suitable ferrite cores for the current transformer

90 degree options for the B-field loop probe and E-field capacitive probe (on E-field probe snap off – scrape copper on each side of slot before you snap off the end to enable soldering)

 

A Bag of Water.

This is a very useful analogy to use when considering an EMC emissions problem, particularly true for radiated emissions in the (often problematic) 30MHz to 1GHz band.

 

Lets get squeezing.

Many of you will have experienced this before. Making a change to an emitting structure inside the equipment by changing the electrical connection between two points results in some emissions going down and some going up.

radiated emissions plot

Then you make another change and this has the opposite effect.

This is like squeezing our bag of water. We can move the water around in the bag much like we can emissions around in the spectrum. The harder we press down in one area, the more it pops up in another.

Emission goes up.

Emission goes down.

 

Reducing the volume

But unless we reduce the amount of water in the bag we will nearly always have a problem. The water is incompressible and it just finds new places to appear.

To achieve this in an EMC context we need to reduce the overall energy in the system.

This could be achieved either by keeping the energy controlled on a PCB away from the radiating structure or by adding lossy components (filters, ferrites, etc) to reduce the amount of energy coupling into the radiating structure.

Changing grounding and bonding within a system without reducing the energy is going to be an exercise in frustration and probably wasted time. Better to address the problem at source where possible.

 

Caveats inbound

There will always be a requirement for us to have to try and achieve the goal of “shaping” our bag of water to fit the radiated emissions limits.

A good example is a manufacturer that has already built a production run of units and needs a quick fix to get them onto the market.

Whilst this is often achievable, there are often significant rework / modification costs involved.

There is also the question of repeatability and consistency. If small changes in bonding of parts can make a large difference to emissions, how can you guarantee that each unit will be compliant? Testing multiple samples can help. As can having good production inspection points during the manufacturing process.

But common mode noise is a slippery customer and these kind of fixes should only ever be considered as temporary pending design changes to address the root cause of the issue.

 

A small plug.

Help is available.

We are really good at this kind of work

We’ve been through the cycle many many times with many many different products.

Using Unit 3 Compliance to help with your emissions problems gets you access to our years of accumulated experience.

Our on site test lab allows us to have a rapid cycle time between analysis of a problem on the bench, developing a fix, and testing in the chamber.

 

Hope this was interesting!

James

Low Frequency, Common Mode, Conducted Emissions

Here is an interesting problem I had working on piece of industrial equipment. The customer had some conducted emissions failures at another EMC lab and needed some help resolving them.

The lessons from fixing this problem was that the first thought is not always the correct one, and that sometimes, all you need is a bit of green-and-yellow earth wire!

 

Outline

A block diagram of the system is shown below with the major components shaded.

An industrial power supply feeds power to the controller (a custom PCB connected to a Raspberry Pi) and to the power measurement board (measures the power consumed by the load).

 

 

Conducted emissions on both the Ethernet port and the AC mains port on the power measurement board were both dominated by a low frequency hump around 700kHz.

 

AC Mains

Ethernet

Notice how the shape or profile of the emissions is almost identical. To my mind, this points towards a single component in the system causing the same noise to be seen everywhere.

 

Simplify First

The first thing I wanted to do was to simplify the test setup as much as possible. I replaced the industrial power supply (often designed for Class A emissions performance) with my trusty Thandar TS3022S adjustable linear bench supply.

The idea here was to eliminate the industrial power supply from my inquiries.

 

 

Wow, what a big difference!

 

So the conclusion here is that the industrial power supply DC output is very noisy, that this noise is propagating through the system, and manifesting as conducted emissions on the outputs via a variety of coupling paths.

 

Differential Mode Filtering

Because conducted emissions noise in this lower frequency range tends to be differential in nature (+ve relative to -ve), my first thought was to implement a differential mode filter on the output of the power supply.

 

 

I’ve got a little filter prototype board that I use in situations like this. This pi filter was made up from two Panasonic FC series 470uF, 25V on either side of a Wurth 33uH iron powder inductor.

 

 

Unfortunately it did nothing to the emissions!

 

Could it be Common Mode?

This sounds like a obvious question to ask in hindsight. Most EMC problems are common mode in nature, I’m just used to thinking about LF conducted emissions as a differential mode problem.

Let’s try a common mode mains filter on the output of the power supply to see if this is indeed the case.

 

 

That’s much better! It looks like the problem was common mode noise after all.

 

This Time It Was Actually A Good Idea…

Common mode noise in this instance is current on both the DC output lines together. But, as I point out in one of my talks, current flows in a loop and always returns to the source. So where is this common mode current returning to? What is it’s reference?

Our common mode emissions measurements are being made in relation to the metalwork of our screened room test setup which is connected to the AC mains Protective Earth (PE).

The AC mains line to each LISN contains a PE connection and, inside the LISN, this is connected directly to the floor of the chamber.

Logically then, connecting the DC negative to the PE on the power supply will provide a shorter path for this common mode noise from the power supply.

 

 

Will this have the desired effect on emissions?

Yes. Yes it does.

AC Mains

Ethernet

 

Conclusion

Ooooooh, bloomin’ common mode noise. Not just for the higher frequencies but lower ones too!

This was a fun half day project fixing this particular problem. Much nicer to be able to recommend a low cost cable assembly than £$€ 20 worth of filter block.

If you’ve got any EMC problems then give me a call, I’d be happy to help.

 

 

dc dc converter emissions before and after with notes on sources

Li-Ion Battery Charger DC/DC Converter – Radiated Emissions Problem Solving

I had a challenging EMC problem solving project in the lab this week.

A customer making a miniaturised 4 cm^3 buck-boost DC/DC converter for Li-Ion battery charging was having radiated emissions issues. The small size meant that adding common mode chokes to filter the input and output connections wasn’t practicable so a more in depth investigation was required.

 

How bad is it?

Here are the emissions for the EUT without any modifications. The green reference trace is the AC/DC mains power supply being used to power the EUT. It is failing the Class B limit (blue) by some margin.

unmodified dc dc converter radiated emissions

Initial Isolation and Investigation

To investigate the emission radiation source (not the cause yet), I placed large clip on ferrite cores around the DC input cable and the battery output cable to reduce emissions directly from the cables.

 

dc dc converter with ferrites on cables

This improves some of the frequencies but not all of them. If the radiation was entirely cable related then this would have dropped the emissions significantly. As it hasn’t, we can conclude that the majority of the emissions are coming from the PCB.

The three peaks we’ll focus on are 180MHz, 300MHz and 500MHz.

Next step is to turn on the spectrum analyser and break out the near field probes. I’ve got a selection of commercial and home made probes but the ones I keep coming back to are the give-away probe cards that I have on my exhibition stand at trade shows.

 

Switching Noise Investigation

The location of the emissions for the 180MHz and 300MHz emissions was initially puzzling. Mostly it was centred around the drain of Q1. If we consider the operation of the circuit, Q1 is turned on permanently in boost mode with Q3 acting as the normal switching element and Q4 acting as a synchronous rectifier. Where is this switching noise coming from?

 

buck boost in boost mode

 

Those of you familiar with synchronous switching converter operation will be shouting at the screen right now. Of course, the answer is bootstrapping.

The high side N channel MOSFETs Q1 and Q4 need a gate voltage higher than their source voltage + their threshold voltage to turn on. In this kind of circuit, this voltage is derived from the switching node via a bootstrap circuit.

This explanation on bootstrap circuit operation from Rohm saves me from re-inventing the wheel.

Even though Q1 is nominally on all the time it still needs to perform a switching operation with Q2 to charge up the bootstrap capacitor powering it’s gate driver circuit.

Checking the datasheet, this switching operation takes just 100ns. That’s very fast indeed and explains the source of our switching noise!

The same bootstrap operation is happening to provide the drive voltage for Q4 but because the boost node is continuously switching this voltage is being provided without such a short switching event.

Due to space constraints it wasn’t easy, but I managed to get the microscope out and modify the board to accept a small but high current ferrite bead in series with Q1 drain.

 

 

500MHz Emission

It didn’t take long to narrow down the 500MHz emissions to the boost output diode D1 with a large amount of ringing on the cathode.

The interesting thing about this diode is that it is only conducting for a very brief period in the dead-time between Q3 turning off and Q4 turning on. Dead time between these parts is set at 75ns, again a very short time period. Good for reducing switching losses, disadvantageous for EMC emission.

 

dead time turn on for the parallel boost diode

 

The part selected for this was a slightly electrically over-rated 40V 1A, SMB packaged part with a reasonable capacitance. Switching 1 amp of current through this part for only a brief period of time before shorting it out and discharging the diode capacitance was causing the ringing to occur.

A ferrite bead was added in series to damp this as the customer wasn’t too keen to head down the rabbit hole of investigating specifying a lower capacitance rated new diode or looking at whether the diode could have been removed altogether at the expense of slightly higher power dissipation in Q4.

Interestingly, this is what the emissions looked like with the diode removed but still with the lower frequency emissions present from the input transistor drain. Note the wideband reduction in emissions above 300MHz.

 

dc dc converter emissions plot with the parallel diode d1 removed from the circuit

Solution

With both of the ferrite beads in place the emissions profile of the EUT was reduced to meet the Class B limits. With more time the peak at 160MHz could be investigated and further reduced but project time pressures and the customer understandably wanting a “good enough” result meant we concluded this investigation here.

 

dc dc converter emissions before and after with notes on sources

The end.

DC/DC converters are often provide a challenging EMC opponent when it comes to radiated emissions. I was glad of the opportunity to work on this project and provide a successful result for the customer. This is the kind of work that I love.

The advantage of being an EMC-consultant-with-a-test-lab combined is that this kind of work can be compressed into hours of work rather than days/weeks oscillating between your lab and the test lab. Problems Fixed Fast!

I hope you found this piece useful, get in touch via the usual channels if you have any questions.

Cheers,

James

 

 

 

HDMI? More like HDM-WHY? Thoughts on Cable Shield Grounding

Ladies and gentlemen, I present this week’s episode of “Crimes Against Cables”

 

Example 1: “I had some leftover components to use”

I’ve seen plenty of interesting EMC “solutions” over the last several years to deal with radiation from cables.

A common one is to separate the shield ground from the signal ground with some combination of components (beads, capacitors, resistors). This approach appears to be particularly common on industrial touch screen display modules for some reason.

poor USB cable grounding suggestions

 

This is (in 99% of cases) a bad idea. I’m not sure what you are hoping to achieve by this and, probably, neither are you 😉

In fact I dedicated a small part of a recent talk to discussing grounds and grounding – you might want to check it out.

 

Example 2: How to Break a Shield

Another notable poor example was an otherwise well crafted piece of military equipment. Shielded connectors and cables all over, it looked like it would be survive some serious electromagnetic abuse (as anything being tested to MIL-STD-461 should).

However,

insulating plastic insert bad idea

insulating plastic insert cross section detail

 

This ends up being not only an emissions problem but an immunity one as well as the cables are just as capable of conducting noise into the shielded case.

This sort of thing can be solved with something like an EESeal type component or by a secondary external screen over the entire assembly.

 

Example 3: Plastic Fantastic

I’ve even seen ferrite cores that were just a moulded plastic lump to appear like cores. Maybe it was a “special” plastic? I never found out, it didn’t help the emissions either.

vga cable ferrite just plastic

But this next one was a first even for me.

 

Example 4 – The Strangest Decision Yet

I was performing a full set of EMC tests to EN 55032 and EN 55035 for a customer. The product had a HDMI interface so obviously there were radiated emissions problems.

The first step as a diagnostic was to use some copper tape to make a connection between the connector shell and the metal back plate – the anodised chassis and EMI gasket material provided was not making a good contact.

This didn’t help so I buzzed the connection with the multimeter to make sure I had some continuity and… nothing.

No connection between the connector shell and PCB Ground.

OK, so there must be a capacitor in series with the shield connection. Fetch the capacitance meter and… 1.2pF.

The board designer had neglected to connect the shield of the HDMI to PCB Ground. It’s a new one for me!

The addition of copper foil to bridge the connector pins to nearby solved the emissions problem but left me wondering why someone thought that was a good idea.

 

I’m going to leave you with this closing thought:

I’ve yet to come across an EMC problem where floating or not connecting a shield ground has improved the situation.

 

 

 

ESD Latch Up Behaviour in Diodes Inc. Power Switch Parts

A new customer came to me with their product that was having problems during testing at another laboratory. There were radiated emissions problems (mostly solved with improvements to the ground plane scheme on the PCB) and a very interesting (and challenging) ESD problem which I’ll cover in this blog.

Here was the device exhibiting the problem, a Diodes Inc AP22802AW5-7 “power distribution load switch”. Input VBAT from a stick of AA batteries, SW_PWR from a rotary switch, and output to the rest of the circuit.

Problem outline

The ESD problem was described by the customer:

The EUT stopped working when 4kV contact discharges were applied on discharge point shown. I removed the batteries and I put them [in] again and there was not any response from the sample (no otuput and the green LED remained OFF).

[A second sample] was then tested with the same result, although this time not on the first discharge

Upon inspection both devices had failed due to the load switch (AP22802AW5-7Diodes), with one failing open and one failing short and both becoming very warm.

ESD diode placed on input and output of load switch (with no effect)

ESD diodes placed on all [discharge points] (with no effect)

ESD diode places on VCC close to pullup resistors for [discharge points] with no effect

First thing first was to get the product set up on the ESD table (with a bit of added blur to protect the innocent).

It was very easy to re-create the problem observed at the original test lab with the second contact discharge to the EUT exposed contact point causing the unit to shut down.

In each case, the power switch was failing low resistance from IN to GND. The initial theory was that the device was being damaged by the high voltage punching through the silicon layers leaving a conductive path.

 

Eliminate the possible

I made a series of experiments to determine the coupling path into the problematic device. Working on the principle that, because of the 15cm distance between discharge point and problem device, that conduction might have been the problem.

  • Capacitors on Vin and EN
  • plus disconnect EN line
  • plus ferrite beads and capacitors on Vin, Vout and EN
  • plus local TVS diodes on pins of device
  • plus ferrite beads in series with [EUT input] lines

Whilst none of these experiments were successful they certainly helped eliminate conduction as the coupling path.

Because of the very high frequency content of the ESD pulse, capacitive coupling is likely going to be the dominant coupling method. Whilst it could couple into the device directly, there was more opportunity for the pulse to couple into the traces connected to the device first. Filtering the inputs eliminates two coupling possibilities

 

Change of sample

The PCB was starting to get a bit tired from the repeated hot air SMT de-soldering and re-soldering so I swapped to another supplied sample. To be able to operate the unit out of the casing I swapped to a linear DC bench supply instead of the AA batteries.

This proved to be an interesting mode as it allowed me to kill the power quickly. The next set of experiments were in an attempt to reduce the effect of capacitive coupling to the problem device.

  • Improved ground stitching / connection
  • Changing supply voltage
  • Indirect HCP discharge – not to EUT but to the Horizontal Coupling Plane albeit with a vertical ESD gun to increase capacitive coupling to EUT.
  • Reduction of coupling into Vin terminal by removing components and copper
  • Addition of copper foil shield over the top of the device

 

Failure mode discovery

Setting the current limit on the DC supply to a fairly low value (about 20% higher than nominal current draw) was a good idea.

When applying the ESD strikes the supply went into foldback as the EUT power input went low resistance. I discovered that quickly turning off the power and then turning it back on effectively reset the failure mode of the device. This proved to be repeatable over several discharges: zap – foldback – power cycle – EUT OK.

What silicon component behaves like this? A thyristor.

This is a phenomena known as “latch up” where the parasitic thyristor structure present in the CMOS process fires due to over voltage… such as an ESD strike for instance!

Because the device is only small the power dissipation caused by the battery short circuit current is enough to “pop” the device through overheating.

 

Out of circuit testing

Whilst it doesn’t get used very often, my Sony Tektronix 370 curve tracer is perfect for testing components like this.

(not mine, picture From CAE Online)

Here’s the VI curve of an undamaged device. It’s a bipolar voltage between VIN and GND. On the left of centre is the standard forward biased body diode. On the right is the reverse biased breakdown of around 8V.

Now for a damaged device. In this case the current changes quickly for a small applied voltage and there is no non-linear characteristic. Essentially, a short circuit.

Turning up the maximum voltage that the curve tracer can apply and dialling down the series impedance allowed me to simulate the over voltage fault condition and create a latch up condition. This latch up wasn’t permanent due to the bipolar sine wave nature of the curve tracer applied voltage.

However turning up the voltage enough to cause excess power dissipation inside the device did result in the same failure mode using the curve tracer.

 

Summary

I have never encountered a device that is this unusually sensitive to ESD events before. A nearby 2kV discharge on the PCB top layer ground plane was enough to cause the latch up condition.

I noted in the report to the customer that this device had been changed to “not recommend for new designs” by Diodes Inc. I wonder if they identified this condition in the device and withdrew it for that reason.

The customer resolved the issue by replacing the device with a different part and we all lived happily ever after.

The end.