The positive rail is at 4 volts. The boost rail reads 3.54 volts, one diode drop, as I expect.
It's been a long road to this point. I checked the naked circuit for shorts between power and ground, that the capacitor was an open circuit, the inductor a short to the power line, and that the resistors were in fact 10K to ground, 820KOhm, and 15kOhm. I applied fresh flux to the pads and got a 'fresh' chip from the desiccant-sealed bag. I used a thermocouple on the chip to monitor the temperature as I soldered it in, being careful to hold the hot air nozzle a few inches from the chip for a minute to warm everything to 100C, then bringing the air gun closer and the temperature up to ~200C, removing the heat as soon as I saw the solder flow. I rechecked the connections for shorts. This is the third attempt today.
It's been a long road to this point. I checked the naked circuit for shorts between power and ground, that the capacitor was an open circuit, the inductor a short to the power line, and that the resistors were in fact 10K to ground, 820KOhm, and 15kOhm. I applied fresh flux to the pads and got a 'fresh' chip from the desiccant-sealed bag. I used a thermocouple on the chip to monitor the temperature as I soldered it in, being careful to hold the hot air nozzle a few inches from the chip for a minute to warm everything to 100C, then bringing the air gun closer and the temperature up to ~200C, removing the heat as soon as I saw the solder flow. I rechecked the connections for shorts. This is the third attempt today.
The 'naked' pads before installing the DRV8662 I connect a jumper wire to the power supply pin. I clip the voltmeter lead onto the boost rail. Holding my breath, I plug the other end of the jumper onto the 'enable' pin to turn on the boost converter. The voltmeter flickers. Something moves on the board. I pull the jumper but it is too late. The voltmeter drops to zero. I smell smoke.
The 1Ohm voltage-sensing resistor on the main power rail now has a dimple in it. It glows red and puffs smoke when I touch the enable pin. It cracks in two when I desolder it. The boost rail reads 2.6 volts. The chip is dead.
The 1Ohm voltage-sensing resistor on the main power rail now has a dimple in it. It glows red and puffs smoke when I touch the enable pin. It cracks in two when I desolder it. The boost rail reads 2.6 volts. The chip is dead.
The Texas Instruments DRV8662 is a "Piezo Haptic Driver with Integrated Boost Converter" used to create high voltage pulses to activate the crystal that makes your phone vibrate. It replaces a transformer the size of a pinky knuckle. It retails for ~$5, going down to $2.50 in bulk. Presumably that's a little less than the circuit from discrete parts, plus space and weight savings. If you're an engineer, your corporate parent pays your salary to save those cents.
The current 'Tabata timer' (lower left) I'm using it because we are building something small and loud. We (The Rise) do workouts outside, in the middle of noisy NYC. We often use a Tabata-style interval of 20 seconds work, 10 seconds rest. Early in their experience, the founders of the group Dave Johnson and Joseph Mullins discovered that this style demands prompt and precise timing. They also discovered that they wanted to coach and participate, not count seconds and blow whistles. A mobile phone app was not loud enough. And we don't have cars in NYC, so was hard to haul a boxing-style timer and 120V AC power inverter/battery to our workouts. So they ended up with a box that Dave built, out of an Arduino micro-controller board, the loudest Radio-Shack piezo-buzzer he could find (it made a difference, particularly as the group grew), and a bunch of AA batteries.
The project I'm working on is the next generation of the box - smaller, with the ability to play different types of sounds, adjust timing, recharge via a USB connection, etc. One design choice we made was to move away from a fixed siren circuit, and instead use an audio amplifier driven by a microprocessor playing sampled sounds.
The DRV8662 is an audio amplifier with a kick. Instead powering the amplifier directly from the battery, the chip includes a boost converter to raise the voltage to a hazardous 105V. The high voltage is necessary to cause the piezoelectric element in our buzzer to flex and create a loud sound.
The converter works by repeatedly shorting an inductor to ground, then opening the circuit, causing a high voltage spike that charges a capacitor. An mechanical analogy is to take a garden hose, turn it all the way on to start the water running, then abruptly shut the valve. The high voltage spike is like the resulting 'water hammer' from the momentum of the water in the hose.
One important distinction of the DRV8662 boost converter is that most commonly available converters will operate with a 2-3x voltage ratio (e.g. converting a 12V input to 24 or 36V). In contrast, the DRV8662 operates at nearly 30x ratio - allowing us to use a single lithium polymer cell to reach the high voltage we want.
A second feature that differentiates this from simpler high voltage circuits used for EL displays and lights is that the oscillator is regulated through a feedback circuit which monitors the boost voltage. It increases and decreases the oscillator frequency to maintain a stable voltage even under significant load - such as the 3kHz siren waveforms we were playing.
Finally, the converter is integrated with the audio amplifier in a single small (4x4mm) package, and also requires relatively few components outside the chip. I could try to build my own converter+amplifier circuit, but TI has done this for me already!
I checked everything, but it still didn't work. What is going on in my circuit?
Boost converters are finicky. I've burned out TI's previous boost-converter based amplifier chip simply by connecting an oscilloscope probe to the output. Because the circuit contains an oscillator controlled by a feedback circuit, anything that disturbs the feedback circuit can potentially cause it to become unstable. If the feedback circuit is unstable, it could go two ways.
Why was my circuit behaving this way? I had done my best to check for shorts, picked conservative component values, taken extra care during soldering, etc. I was running out of chips, had searched all of the message boards, and getting frustrated.
How many times have I puzzled over a circuit, suspicious of manufacturing defects, strange alien phenomena, or an act of God preventing it from working, only to check the datasheet and find that I had connected the pins backwards?
Armed with a hypothesis about the feedback circuit, I looked again through the datasheet. In the 'Thermal/layout considerations' section, after a long paragraph on the importance of the thermal footprint, I read, with renewed interest and suspicion:
The project I'm working on is the next generation of the box - smaller, with the ability to play different types of sounds, adjust timing, recharge via a USB connection, etc. One design choice we made was to move away from a fixed siren circuit, and instead use an audio amplifier driven by a microprocessor playing sampled sounds.
The DRV8662 is an audio amplifier with a kick. Instead powering the amplifier directly from the battery, the chip includes a boost converter to raise the voltage to a hazardous 105V. The high voltage is necessary to cause the piezoelectric element in our buzzer to flex and create a loud sound.
The converter works by repeatedly shorting an inductor to ground, then opening the circuit, causing a high voltage spike that charges a capacitor. An mechanical analogy is to take a garden hose, turn it all the way on to start the water running, then abruptly shut the valve. The high voltage spike is like the resulting 'water hammer' from the momentum of the water in the hose.
One important distinction of the DRV8662 boost converter is that most commonly available converters will operate with a 2-3x voltage ratio (e.g. converting a 12V input to 24 or 36V). In contrast, the DRV8662 operates at nearly 30x ratio - allowing us to use a single lithium polymer cell to reach the high voltage we want.
A second feature that differentiates this from simpler high voltage circuits used for EL displays and lights is that the oscillator is regulated through a feedback circuit which monitors the boost voltage. It increases and decreases the oscillator frequency to maintain a stable voltage even under significant load - such as the 3kHz siren waveforms we were playing.
Finally, the converter is integrated with the audio amplifier in a single small (4x4mm) package, and also requires relatively few components outside the chip. I could try to build my own converter+amplifier circuit, but TI has done this for me already!
I checked everything, but it still didn't work. What is going on in my circuit?
Boost converters are finicky. I've burned out TI's previous boost-converter based amplifier chip simply by connecting an oscilloscope probe to the output. Because the circuit contains an oscillator controlled by a feedback circuit, anything that disturbs the feedback circuit can potentially cause it to become unstable. If the feedback circuit is unstable, it could go two ways.
- The oscillator slows down, increasing the size of the voltage spike. The boost voltage increases until it exceeds the voltage rating of the capacitor, diode, or something else in the circuit.
- The oscillator stops (or in our case, never starts), causing a short circuit through the inductor and a large current to flow through the circuit.
Why was my circuit behaving this way? I had done my best to check for shorts, picked conservative component values, taken extra care during soldering, etc. I was running out of chips, had searched all of the message boards, and getting frustrated.
How many times have I puzzled over a circuit, suspicious of manufacturing defects, strange alien phenomena, or an act of God preventing it from working, only to check the datasheet and find that I had connected the pins backwards?
Armed with a hypothesis about the feedback circuit, I looked again through the datasheet. In the 'Thermal/layout considerations' section, after a long paragraph on the importance of the thermal footprint, I read, with renewed interest and suspicion:
Another key layout consideration is to keep the boost programming resistors (R1 and R2) as close as possible to the FB [feedback] pin of the DRV8662. Care should be taken to avoid getting the FB trace near the SW [switch] trace.
Example PCB layout from the DRV8662 datasheet
It also provided a recommended layout, used in the evaluation board. Since I had been just learning my PCB layout program, I thought it would be better to lay things out myself, rather than copy someone else's layout! And reading PCBs was still fairly intimidating.
Upon studying the layout, I began to understand that in this layout, the resistors R1 and R2 are connected to the feedback pin on the left-hand side and to the boost pins by a trace on the backside, which also stays away from the switch pins. In my layout, I had placed R1 and R2 somewhere south of the inductor, and connected them to the feedback pin by a long backside trace, because it had been getting crowded on the left side. Could this be the culprit? It was puzzling, because in the early breadboard stages I had gotten a chip to work, and that had much longer traces. But there was only one way to find out.
It took me a couple of hours to revise the layout and then go through my in-house PCB manufacturing process. I skipped the power section of the board and concentrated on the DRV8662, installing the passives, checking the circuit, and finally soldering the chip in place.
Finally, I powered up the circuit. The input was at 4 volts, the boost output at 3.58V. Crossing my fingers and listening for the telltale 'click' from the power supply switching into current-limiting mode, I connected the enable pin.
The voltmeter read 74.5V - exactly as I had programmed. I quickly disconnected the enable pin, checked for any smoke, and connected it again. Same reading! Sigh of relief.
I still don't completely understand why the feedback circuit is so sensitive to the position of the resistors. The best guess that I have is that the feedback resistors form a voltage divider for the high-voltage signal - and that a smaller signal is more susceptible to interference from the switch traces. Hence, put the voltage divider near the feedback pins, and be careful not to get too close to the switch traces. But I would have expected this to affect the final voltage, not simply prevent the circuit from starting in the first place. And how sensitive is it? How far do I need to move the resistors before they cause me trouble? EDIT: I heard back from the application engineer at TI in the haptics forum within a few hours of posting a question about this issue - he confirmed my theory. Thanks, TI!
Regardless of the cause, this parameter is critical in the PCB layout. And I reminded myself, I'd rather find it out now, rather than when I've placed an order for a thousand PCBs!
This was my first major circuit to come online - and I had struggled with it for probably two weeks. But I've learned a lot - how to check a naked circuit, surface mount soldering and temperature control, in-house PCB manufacturing, debugging a circuit, and the finer points of boost converters, and reading datasheets carefully. A significant iteration.
Thanks for reading!
Postscript: I've since also found TI's next generation chip - the DRV2667 has an I2C interface for controlling gain, and is pin-compatible with DRV8662.
Upon studying the layout, I began to understand that in this layout, the resistors R1 and R2 are connected to the feedback pin on the left-hand side and to the boost pins by a trace on the backside, which also stays away from the switch pins. In my layout, I had placed R1 and R2 somewhere south of the inductor, and connected them to the feedback pin by a long backside trace, because it had been getting crowded on the left side. Could this be the culprit? It was puzzling, because in the early breadboard stages I had gotten a chip to work, and that had much longer traces. But there was only one way to find out.
It took me a couple of hours to revise the layout and then go through my in-house PCB manufacturing process. I skipped the power section of the board and concentrated on the DRV8662, installing the passives, checking the circuit, and finally soldering the chip in place.
Finally, I powered up the circuit. The input was at 4 volts, the boost output at 3.58V. Crossing my fingers and listening for the telltale 'click' from the power supply switching into current-limiting mode, I connected the enable pin.
The voltmeter read 74.5V - exactly as I had programmed. I quickly disconnected the enable pin, checked for any smoke, and connected it again. Same reading! Sigh of relief.
I still don't completely understand why the feedback circuit is so sensitive to the position of the resistors. The best guess that I have is that the feedback resistors form a voltage divider for the high-voltage signal - and that a smaller signal is more susceptible to interference from the switch traces. Hence, put the voltage divider near the feedback pins, and be careful not to get too close to the switch traces. But I would have expected this to affect the final voltage, not simply prevent the circuit from starting in the first place. And how sensitive is it? How far do I need to move the resistors before they cause me trouble? EDIT: I heard back from the application engineer at TI in the haptics forum within a few hours of posting a question about this issue - he confirmed my theory. Thanks, TI!
Regardless of the cause, this parameter is critical in the PCB layout. And I reminded myself, I'd rather find it out now, rather than when I've placed an order for a thousand PCBs!
This was my first major circuit to come online - and I had struggled with it for probably two weeks. But I've learned a lot - how to check a naked circuit, surface mount soldering and temperature control, in-house PCB manufacturing, debugging a circuit, and the finer points of boost converters, and reading datasheets carefully. A significant iteration.
Thanks for reading!
Postscript: I've since also found TI's next generation chip - the DRV2667 has an I2C interface for controlling gain, and is pin-compatible with DRV8662.
RSS Feed