Sleep current 24 µA instead of 2 µA: six weeks of battery life, gone
A sensor node designed for two years on a 2200 mAh cell died in about six weeks in the field. Sleep current measured 24 µA against a 2 µA budget. The 22 µA was located by subtraction: a microamp meter in series with the battery, then one block disabled at a time. · 5 µA: LDO quiescent current, from the quick pick rather than the 1 µA part · 6 µA: a 10 k pull-up on an input held low permanently, 3 V across 10 k · 4 µA: ADC reference left enabled after initialisation · 3 µA: brown-out detector in full-accuracy mode · 2 µA: flash standby, which is not standby without the command · 2 µA: MCU deep sleep, the one line item that matched the budget Changes: 1 µA LDO, 1 MΩ pull-up, reference and flash disabled in the sleep path, low-power BOD. Sleep current measured 1.8 µA, and the node now calculates to roughly 20 months. The work is a spreadsheet rather than a chip selection: every block, its sleep current, multiplied by 8,760 hours, compared against the cell that fits the enclosure. The meter matters too: a 10 MΩ DMM in series with a 3 V cell adds about 0.3 µA of error, which is 15% of the target at this level. #LowPower#IoT #SleepCurrent
A 40 cm ribbon turned a 25 MHz clock into a three-ring oscillator
A 25 MHz clock from an MCU to a peripheral over a 40 cm ribbon cable looked clean at the source and unusable at the receiver: 4.1 V overshoot on a 3.3 V rail, and three visible rings before it settled. Reflection. Two attempts failed before the fix. A 100 pF capacitor at the receiver slowed the edges and made the ringing slightly worse. Reduced output drive strength helped the overshoot but left setup time marginal at the receiver. What worked: · 33 Ω series resistor at the source, matching the driver's output impedance to the trace and terminating the first reflection · a ground return adjacent to the clock in the ribbon, which had been running beside two data lines with no return · ribbon shortened from 40 cm to 22 cm · clock moved away from the regulator's inductor in the connector pinout Result: 3.5 V peak, one small ring, clean eye at 25 MHz. The return path is part of the signal. On a ribbon sharing one ground wire across six signals, return inductance is high, and source termination cannot compensate for it. The resistor value was not guessed either: 33 Ω brought the driver's 25 Ω output impedance up to the 50 Ω trace, while 22 Ω left a visible step in the edge. #SignalIntegrity#Reflection #SeriesTermination
Datasheet efficiency 92%, measured 82%. Where the missing 10% went
A 12 V to 5 V synchronous buck specified at 92% efficiency measured 82% on the board at 3 A. Loss was split into parts instead of being blamed on the controller: four-wire meters at input and output, plus the regulator's own loss estimate. The IC reported about 0.4 W dissipating; total loss was 3.3 W, leaving 2.9 W outside the chip. The missing pieces: · inductor DCR of 28 mΩ at 3 A: 0.25 W, plus 0.4 W of core loss at 500 kHz · an inductor rated 2 A, saturating slightly and heating, which raised its DCR further · forced PWM at light load, so the low-side FET kept switching when it did not need to · 500 kHz instead of 300 kHz: about 1 W of switching loss traded for inductor size · a 10 mΩ current-sense resistor: 0.09 W absent from every model Changes: 3 A inductor with 18 mΩ, 300 kHz, light-load mode enabled, and a sense method that adds no series resistance. Efficiency reached 89% at 3 A and 92% at 4 A. Most efficiency gaps traced so far were magnetics, frequency and measurement, not the controller. The measurement itself needs care: a drifting input voltage produces a gap that looks like loss, so input and output are logged at the same instant. #Efficiency #PowerLoss #Measurement
Calculated loss 0.4 W, measured 2.9 W. Same MOSFET, same current
A synchronous buck running 12 V to 5 V at 4 A measured 110 °C at the MOSFETs. Calculated conduction loss was 0.4 W, which does not produce 110 °C. Measured breakdown: · RDS(on) at 100 °C rather than 25 °C: 0.4 W became 0.6 W, since the datasheet value roughly doubles when hot · switching loss at 500 kHz with 20 ns edges: about 1.6 W across both FETs · gate charge at 500 kHz with a weak driver: 0.3 W · dead time of 30 ns: 0.4 W of body-diode conduction · inductor core loss heating the surrounding copper Two changes carried most of it. Switching frequency dropped to 300 kHz, cutting switching loss by about a third at the cost of a larger inductor. Dead time came down to 12 ns with a better driver, halving the diode term. Result: 78 °C at the same load, efficiency from 86% to 91%. Conduction, switching, gate and dead-time losses now get summed before any thermal calculation is trusted. When the sum disagrees with the measurement by a factor of seven, the measurement is right and the list is missing a term. The thermal camera also showed the inductor at 95 °C, heating the local air rather than adding directly to the FET's own loss. #MOSFET#SwitchingLoss #Efficiency
Polling the encoder worked at 300 rpm. At 900 rpm, 20 counts per second went missing
A joint test rig showed about 3° of position drift per minute at high speed and none at low speed. Mechanical slip was the first assumption. The cause was polling. Firmware read the encoder inside the 1 kHz control loop. At 900 rpm with a 1000-line encoder in quadrature, edges arrive at roughly 240 k per second, one every 4 µs. Reading a counter every 1 ms is fine when hardware does the counting. This code read pin states and incremented in software, so every edge that arrived while the loop was busy disappeared. A logic analyser measured 20 missing counts per second at 900 rpm, which matches the drift. Changes: · timer switched to encoder mode, making the count independent of loop timing · 1 kHz snapshot of the counter in place of pin reads · overflow interrupt added, since the counter wraps often at that edge rate · polling build retained for comparison Drift went to zero, repeatability improved, and the loop shortened once the pin reads were gone. Polling is a promise that the loop runs faster than the event being watched. At 240 k edges per second against a 1 ms loop, that promise was off by a factor of 240 #Interrupt#Encoder #Firmware
The board booted at random when touched. The EN pin was floating
A board powered up at random when handled, and sometimes failed to start on reset. Firmware and hardware each blamed the other. The regulator's EN pin had no pull-down: a test point and a via to the expansion header, nothing else. A meter on that pin while a hand moved near the board read 1.8 V, then 0.9 V, then 3.1 V. An undriven input settles wherever nearby charge puts it. Two failures followed: · random power-up, when EN drifted above threshold during handling · about 400 µA through the pin's ESD structure when it floated above VCC with the rail off, matching a handful of dead regulators in one batch Fix: 100 k pull-down at the pin, 1 k series resistor to the header so a miswired cable cannot inject current. Every enable, reset, mode and boot-select pin went through the same check. The rule recorded from it: every input pin needs a defined level when nothing drives it. Internal pull-ups apply only after the MCU is powered, and reset happens before that. The review question for each pin is now fixed: what holds this pin when the board is off, in a bag, or in a hand? The same pass found two more pins in the same condition, both on the expansion header, both invisible in schematic review because a net with a test point looks defined. #GPIO#PullUp #BringUp
Reverse polarity protection that worked — until a 24 V hot plug
Reverse polarity protection on a 24 V input used a P-channel MOSFET with the gate tied to ground through a resistor. A reversed supply was blocked correctly on the bench. A 24 V hot plug into a board carrying a 470 µF input bank destroyed the FET shorted, which is the worst failure mode for a protection device. The scope showed the sequence. The body diode conducts first into the discharged bank, holding full Vds across the FET while amps of inrush flow through the diode. The gate network turns the channel on several hundred microseconds later, into a bank still charging. The drain-source transient passed 40 V on a 30 V part. Changes: · 40 V FET, with margin taken over 24 V plus transients · TVS at the input, sized against the ringing measured at the adapter output during cable arcing · 10 Ω gate resistor and a soft-start network, so turn-on is gradual · inrush limiting on the capacitor bank The same hot plug test at 24 V ran 200 cycles without failure, and the connector arcing that had been eroding the XT60 contacts stopped. Reverse protection against a battery is a DC problem. Against a cable with inductance it is a transient problem, and the real adapter is the test source that matters. #Protection#PMOS #HotPlug
A low-side MOSFET ran at 96 °C. The 10 k gate resistor was the cause
A low-side N-MOSFET switched a 24 V, 3 A load at 100 kHz. Calculated conduction loss was 0.25 W. Measured case temperature was 96 °C, equivalent to roughly 3 W. The gate was driven from a 3.3 V MCU pin through a 10 k resistor inherited from an earlier schematic. Against 2 nC of gate charge, gate rise and fall took 400 ns, and the drain transition took longer. At 100 kHz, 800 ns of transition per cycle is 8% of the period spent in the linear region, with voltage and current on the die at the same time. That is where the dissipation came from. Changes: · gate resistor reduced to 4.7 Ω and a dedicated gate driver replacing the MCU pin · rise time measured at 18 ns, case temperature 52 °C · 100 k pull-down added, since the pin floats during reset and one FET had been observed half-on · snubber across the drain to damp the ringing The conduction calculation was correct and incomplete. Switching loss is the dissipation while the FET behaves as a resistor with voltage across it. A hot FET with low calculated loss points at the gate waveform before it points at the part number. The 10 k had been carried over from an earlier design switching at 1 kHz, where the slow edge cost nothing measurable. #MOSFET#GateDrive #Thermal
A 10 µF ceramic measured 3.1 µF at 5 V. Ripple went from 15 mV to 80 mV.
A 5 V rail showed 80 mV of ripple against 15 mV in simulation. The bank was four 10 µF 0603 X5R 6.3 V capacitors, matching the reference design. On the LCR meter at 5 V DC bias, one capacitor read 3.1 µF at 25 °C. Four parts were delivering roughly a quarter of the nominal capacitance. X5R loses capacitance with both DC bias and temperature. A 6.3 V part on a 5 V rail is already at 80% of rated voltage before anything else happens. Changes made: 10 µF 1206 X7R 16 V replacing the 0603 6.3 V X5R, measured 8.2 µF at 5 V one 22 µF electrolytic in parallel, chosen for stability across bias and for damping the ceramic piezo resonance vendor bias curves filed with the design notes Ripple fell to 14 mV and matched simulation once real capacitance replaced the reel value in the model. Sizing output capacitance from the label leaves a design 30–70% short at operating voltage. The LCR meter at bias is the only measurement that closes the gap. The electrolytic was not a bulk substitute: it held 22 µF at 5 V where the ceramics had given up most of their value, and the pair flattened an impedance peak the ceramics created near 2 MHz.
A 2 A inductor was fine at 25 °C. At 90 °C it saturated at 1.4 A.
A 5 V to 3.3 V buck was designed for a 2 A load with an inductor specified at 2.2 A saturation. Bench tests at room temperature passed. Inside the enclosure at 90 °C, the output fell to 2.9 V on a 1.8 A step. The inductor current on the scope ramped and then went vertical at about 1.4 A. Saturation. The datasheet figure was accurate at 25 °C. Saturation current derated roughly 40% at 90 °C, and the BOM carried only the headline number. What entered the review afterwards: Isat at the actual ambient rather than 25 °C Irms at temperature, since DCR rises about 30% between 25 and 100 °C core loss at the switching frequency, which accounted for most of the extra heat the regulator's current limit, which was what actually protected the FET The replacement was a 4.7 µH part in a larger package, Isat 3.4 A at 100 °C, positioned away from the FET. Output held 3.28 V through the same load step. BOM comments now carry "Isat @ 100 °C", so the next review does not rest on a footnote read at 25 °C. Local air near the inductor also measured about 15 °C above the ambient sensor, which sat in the opposite corner of the board.
What Would You Build for Your Desk? 🛠️
A USB-C power monitor? A tiny weather station? A programmable macro pad? Or a little robot that keeps you company while you work? Small desk gadgets are great projects for experimenting with sensors, displays, microcontrollers, and custom PCBs. Sometimes the best project is the one you build just because you can. #DIYElectronics#DeskGadgets #EmbeddedSystems#MakerProjects
It Worked in the Simulator. Then Came the Real Board.
The power rail looks fine. The firmware compiles. But the hardware still doesn't behave as expected. Maybe it's a wiring issue, a timing problem, or a small detail in the PCB layout.😅 Debugging isn't always the most exciting part of building electronics, but it's where a lot of learning happens. What's a hardware bug that took you much longer to solve than expected? #ElectronicsEngineering#HardwareDebugging #PCBDesign#Maker
From Schematic to Something You Can Hold
A circuit diagram on the screen is one thing. Seeing it become a working board is another. There's component selection, PCB layout, assembly, testing, and usually a few unexpected problems along the way. But that's part of the process. #PCBDesign#HardwareDevelopment #ElectronicsEngineering#Maker
Electronics Can Be More Than Just Useful ⚡
A tiny desk gadget, a custom game controller, or a robot that reacts to its surroundings — hardware projects can be practical, playful, or a bit of both. That's what makes building things with electronics so much fun. What's the most fun electronics project you've worked on? #DIYElectronics#MakerProjects #OpenSourceHardware
Every Great Hardware Project Starts Somewhere 🔧
Before the first LED lights up or the first sensor returns a reading, there's a circuit to design and a PCB to build. From a quick weekend prototype to a fully functional device, every project starts with an idea and a lot of small decisions. What are you building on your workbench right now? #OpenSourceHardware#DIYElectronics #PCBDesign#Maker
Designing Via-in-Pad in EasyEDA
Here's a quick look at designing via-in-pad in EasyEDA.
The idea is simple: place the via directly inside the component pad instead of routing it to a nearby via. This can free up valuable space when you're working with a tight PCB layout.
Check out the video to see the design process in action.
📌 Save it for reference when you work on your next board.
More about via-in-pad:
https://jlcpcb.com/help/article/via-in-pad-pofv-guide
#EasyEDA #PCBLayout #ViaInPad
The Difference Between a Prototype and a Product
A prototype proves that an idea can work. A product requires much more. Reliability testing, manufacturing consistency, cost optimization, and long-term stability all become important. The journey from a working circuit to a real product is where engineering truly begins. #HardwareEngineering#Manufacturing #PCB
Why Design for Manufacturing Matters
A PCB that works in the lab is not always ready for production. Manufacturing constraints such as trace spacing, solder mask, component placement, and assembly process need to be considered early. Good design is not only about performance—it is also about making reliable products efficiently. #DFM#PCBAssembly
Why Current Return Paths Matter
Many engineers focus only on where the signal goes, but the return path is equally important. A poor return path can increase loop area, introduce noise, and create unexpected electromagnetic radiation. Designing the return path is a key part of modern PCB engineering. #EMC#SignalIntegrity
Why Four-Layer PCBs Are Popular
A four-layer PCB is not just a bigger two-layer board. A typical stack-up with dedicated power and ground planes improves signal integrity, reduces noise, and makes routing easier. For many modern electronic products, the additional layers provide a good balance between performance and cost. #PCBManufacturing#PCBDesign
Why Datasheets Save Hardware Engineers Time
Datasheets contain more than electrical specifications. They often include recommended layouts, placement guidelines, thermal requirements, and application examples. Reading the datasheet carefully before designing can prevent many problems during testing and manufacturing. Good engineering starts before the schematic is drawn. #Engineering#Electronics
Why Ground Is Not Just Another Signal
Ground is often treated as a simple connection, but in real circuits it plays a much more important role. It provides the reference voltage and the return path for current. Poor grounding can create noise, instability, and EMI problems. Understanding current flow is one of the biggest steps in improving hardware design skills. #Grounding#Electronics #PCB
Why Component Placement Comes Before Routing--from jlc--
A common beginner mistake is starting routing immediately after importing the schematic. In reality, component placement determines most of the final PCB performance. Power paths, signal flow, thermal areas, and mechanical constraints should be considered before drawing traces. A good layout usually starts with good placement. #PCBLayout#HardwareDesign
Why BGA Packages Change PCB Design
BGA packages provide high pin density and excellent electrical performance, but they also introduce routing challenges. The escape routing from the package requires careful planning, often involving microvias, HDI technology, and advanced stack-up designs. As component density increases, PCB manufacturing capability becomes an important part of the design process. #BGA#HDI #PCBDesign
Why Thermal Design Starts During PCB Layout
Thermal problems are often discovered too late.
Copper areas, thermal vias, component spacing, and airflow should be considered from the beginning of the design process.
A circuit that works electrically may still fail if heat is not managed properly.
Good hardware design considers both electrons and temperature.
#ThermalDesign #PCB #HardwareEngineering
Ripple and Noise Are Not the Same Thing
Power ripple and noise are often mentioned together, but they have different characteristics.
Ripple is usually a periodic variation related to switching frequency or rectification. Noise is typically random high-frequency interference caused by EMI, parasitic effects, or switching transients.
Understanding the difference is important because they require different solutions.
#PowerIntegrity #Electronics
Why Crystal Oscillators Need Special Attention
A crystal oscillator may be one of the smallest parts on a PCB, but it controls timing for the entire system. Long traces, nearby noise sources, and poor grounding can affect clock stability. Keeping the crystal close to the MCU and following layout recommendations from the datasheet can prevent many unexpected problems. #EmbeddedSystems#PCBDesign
Learning Hardware Means Learning From Mistakes
Hardware mistakes are sometimes expensive, but they are also valuable teachers. A wrong footprint, a missing connection, or an unexpected noise problem can teach more than many hours of reading. Every engineer has experienced failed boards, but each failure builds better instincts for future designs. #ElectronicsEngineering#Maker #Hardware
Learning Hardware Means Learning From Mistakes
Hardware mistakes are sometimes expensive, but they are also valuable teachers. A wrong footprint, a missing connection, or an unexpected noise problem can teach more than many hours of reading. Every engineer has experienced failed boards, but each failure builds better instincts for future designs. #ElectronicsEngineering#Maker #Hardware
Why Crystal Oscillators Need Special Attention
A crystal oscillator may be one of the smallest parts on a PCB, but it controls timing for the entire system. Long traces, nearby noise sources, and poor grounding can affect clock stability. Keeping the crystal close to the MCU and following layout recommendations from the datasheet can prevent many unexpected problems. #EmbeddedSystems#PCBDesign