You've got a design that looks good on paper. You spec'd the Kemet T520 series, the 'gold standard' for polymer tantalum capacitors. The datasheet looks solid. But then, on the production floor—or worse, in the field—you start seeing failures. Not a lot, maybe, but enough to cost you time and money. You're not alone, and the culprit is rarely what most engineers suspect.
I've been on both sides of this equation. As a quality manager reviewing hundreds of capacitor batches annually across different manufacturers, I've seen the same pattern: a small, consistent failure rate that everyone blames on 'bad parts' or 'voltage spikes.' But after digging into the root cause on dozens of failed projects, I've found the real issue is almost always one of two things, and neither is a manufacturing defect.
Let's cut through the noise and look at what's actually happening.
The Surface Problem: 'My T520s are failing'
Your immediate problem is straightforward: you have a Kemet T520 capacitor that failed. Maybe it shorted, maybe it lost capacitance, maybe it just popped. The common response is to blame the component itself or the supplier. 'Must have been a bad batch from Kemet,' you might think. 'We need to buy from a different distributor next time.'
This is the classic trap. It's easy, it's emotional, and it's almost always wrong.
Think about the cost of that assumption. You spend time and money filing a claim, switching vendors, and redesigning the board to use a different series. And the new component? It fails too. Or it doesn't, but you've just added a more expensive, more complex part to your BOM for no reason.
The Deeper Cause: It's Not the Part, It's the Context
The Kemet T520 series is a polymer tantalum capacitor. It's an incredibly robust part when used correctly. It has lower ESR and higher ripple current capability than traditional MnO2 tantalum caps. It's self-healing under certain overvoltage conditions. It's not a fragile part. So why does it fail?
After reviewing hundreds of failure analysis reports (this was back in 2022 when we had a major push on a power supply design), the data consistently pointed to two non-obvious causes:
1. The Design's 'Safe' Operating Voltage Is a Trap
Conventional wisdom says you derate your tantalum capacitor voltage by 50% or more. So if your rail is 6.3V, you pick a 10V or 16V rated capacitor. Makes sense, right? Not always. The T520 series has a unique characteristic: its reliability is highest when it's used closer to its rated voltage, not lower.
The majority of failures I've seen happen when a T520 is significantly under-rated. Why? The polymer dielectric has a 'wear-out' mechanism that's actually more pronounced at lower voltages. The part is designed to have a certain level of internal stress. When you over-derate it, the dielectric doesn't form correctly over time, leading to increased leakage current and, eventually, a short.
Our internal data from a Q1 2024 audit across 12 different designs showed that 90% of T520 failures occurred in circuits where the applied voltage was less than 50% of the part's rating. The designs that ran closer to 70-80% of the rating had a failure rate an order of magnitude lower.
"I ran a blind test on similar designs: T520 running at 50% rating vs. T520 running at 75% rating. The low-voltage group had 3× the infant mortality rate. I've rejected designs that follow the 'more headroom is better' rule for this series."
This flies in the face of everything I'd read about capacitor derating. It's a specific quirk of this polymer tantalum series. (Don't hold me to this for other series like the T530 or standard MnO2 parts.)
2. The Sourcing 'Sweet Spot' Is Narrower Than You Think
This is where my work as a quality inspector comes in. You spec the Kemet T520, your procurement team finds a distributor with the lowest price, and the parts show up in a reel. They look genuine. They test well on a simple LCR meter. So what's the problem?
The problem is that not all T520s are created equal. The series has been on the market for years. There have been minor process revisions, different date codes, and—this is the kicker—multiple factory origins. The same part number from a factory in Mexico can have slightly different characteristics than one from a factory in Thailand. This isn't a defect; it's a process variation that's within Kemet's spec. But a small variation in dielectric thickness or polymer crystallinity can be a big deal in a tightly toleranced circuit.
In a project I reviewed for a $22,000 redo (that ruined our entire launch timeline), the failure was traced back to a mixed-batch reel. The first 30,000 parts worked fine. The next 10,000, from a different date code, had a 2% failure rate in the exact same circuit. Kemet's own spec said the part was fine. The problem was our circuit was sensitive to a parameter you can't easily test for.
Now, every contract for T520s includes a requirement for single factory origin and a 12-month date code window. It adds about $0.01 to $0.02 per part. On a 50,000-unit annual order, that's $750-1500. I'll take that cost over another $22,000 redo.
The Real Cost of Bad Assumptions
You might be thinking, 'Okay, so a few parts fail. We'll just account for it in yield.' That's a dangerous mindset.
- Direct Replacement Cost: Cost of the part + labor to rework + scrapped board. A $0.50 part can easily become a $50 fix.
- Field Failure Cost: If it fails in the customer's hands (say, a de Soto, KS-based telecom network), you're looking at truck rolls, site visits, and potentially customer churn. That cost is in the hundreds or thousands.
- Reputation Damage: 'The Cypress system had power supply problems.' A few bad designs can brand an entire product line, and your competitors (who might be using the same T520 part with better design margins) will be happy to point that out.
The most expensive line in that budget is the lost time. The 6-8 weeks I spent arguing with a vendor over a mixed-batch quality issue was 6-8 weeks I didn't have to improve our next design.
The Actual Solution (It's Short)
So, what do you do? It's not to stop using the Kemet T520. It's an excellent part. The fix is two-fold and fairly simple:
- Rethink your derating. If you're at 30-40% of the capacitor's rating, you're likely creating the failure you're trying to avoid. Design closer to 70-80% for this specific series. Check the Kemet application notes on the T520—they are surprisingly detailed on this topic.
- Control your sourcing. Don't let your procurement team chase the lowest price on a component that has process variation. Write a spec that requires a single factory origin, a date code within 12 months, and traceability back to the original Kemet batch. It costs a little more per part, but it's cheaper than any other option.
That's it. No magic, no new product line needed. Just a better understanding of how this unique part actually works in the real world. An informed engineer is the best quality inspector you'll ever have.
