07/18/2019 at 23:22 •
The boards arrived and I fabricated one. See photos. Flexible PCB from OSHPark is rather pretty.
It folds nicely. I bent the tabs that fit in the slots at right angles. That makes them stay in the slots without glue. You can bend the PCB severely, at less then the radius recommended (about 2mm) and the bend will more or less stay. In my case, there is no circuitry on that portion.
I gingerly bent the sides down around a 1mm rod (an impromptu mandrel.)
I spent much time debugging, only to find that I had oriented and installed the solar cells backwards. Solar cells are diodes. But the current they generate is in opposite direction to the current you would put through a normal diode. That is, in the opposite direction of the arrow part of the diode symbol. Remember, I am using photodiodes as solar cells. So the base of the arrow is the positive side of the diode as a current/voltage source.
I usually draw an "orientation" diagram for all the parts that need orientation during fabrication. But I got it wrong.
And, on the KiCad footprint I made, I omitted a diode symbol on the silkscreen that would have helped me orient the solar cells.
Also, it turns out the footprint I made was not correct. So the part pins barely overlapped the lands. They did reflow, but are likely to suffer more soldering failures than usual. I will need to revise.
Also, the TPS3839 footprint is wrong. I think it should be SOT23-3 (a TI name?) instead of SOT-323? Again, the part pins barely overlapped the lands on the PCB. In a revision, I think I will use a TI "DQN" package, a tiny 4 pin package for a single transistor.
Finally, the earring does turn, but as noted before, it requires bright light: at least outdoor shade with much exposure to bright sky. A better design would use a little bit larger solar cells. But again, they are not easily available.
06/28/2019 at 13:20 •
I was suprised to learn how expensive flex PCB is. In my experience, roughly five times more expensive.
My typical, small, rigid board costs on the order of a few dollars for three. My first flexible PCB board was over twenty dollars for three (although it was about twice as large as my typical board.)
One issue is that a flexible board is more likely to have wasted space, that is, the bounding box (for which you are charged) is larger than the actual board area. A flexible PCB is liable to have dangling parts, or extensions, where you might ordinarly use wire cables. Such things will increase the bounding box.
The board of this project had extensions in four directions. The bounded area was fifty percent waste: 6 areal units of waste, 6 areal units of actual board.
If your use case is: just need the lightness, don't need the extensions, then the cost differential won't be as great.
You can panelize your board (tile a plane, bin pack) to reduce the waste.
06/28/2019 at 13:08 •
A few notes about how long it takes to get your flexible PCB. Your experience may vary.
OSHPark aggregates your design onto a larger PCB for fabrication, then breaks your portion out and mails it to you. The steps are (OSHPark usually emails you a notification for each) 1. assigned to board 2. board sent to fab 3. board received from fab 4. board shipped to you.
My experience with rigid PCB is that the whole process takes about a week to ten days. The longest step is at the fab. Often, my boards (which are tiny) are assigned to a board and sent to fab within one day.
With flexible PCB, the whole process takes about 3 weeks. The longest step is "assigning to board". My first flexible PCB took two weeks to be sent to fab. I suppose that the demand for flex PCB is so much less than for rigid, that it takes longer before a board fills up with aggregated designs. Also, I suppose that most flexible PCB's are small (because they are more expensive, so often only the part of the circuit that needs to be flexible is submitted and the rest of the circuit is implemented on rigid PCB.) That also would mean it takes longer for a larger board to fill up so it can be sent to fab.
05/28/2019 at 20:58 •
The timing circuit (diode Dhold and timing capacitor Chold) determines how long the motor is energized. The voltage on the capacitor Chold and the Vsense network is “held up” even as the voltage on the motor falls. The capacitor Chold discharges only through the operating current of the voltage monitor, say 150nA.
You can use the formula F=A*S/V to calculate how long in seconds before the voltage on Chold capacitance (say 0.047uF) falls to the hysteresis voltage of the voltage monitor (Vth - 50mV) at the current of the voltage monitor (150nA.) The unknown in the formula is S in seconds. V is 50mV (delta V.)
That determines how long the voltage monitor and mosfet energize the motor from the storage capacitor. You can use the same formula to crudely determine how far the voltage will drop on the storage capacitor. The unknown is V (or delta V.)
In practice, rather than do the calculations, I just experiment. I suppose you might also be able to simulate the circuit.
You can use such an excessively large Chold (say 10uF) that the storage capacitor completely discharges. See below for more discussion.
The voltage on the motor is the Vth (1.1V) of the voltage monitor plus the voltage drop of the diode Dhold. I have specified a Schottky diode, with a voltage drop around 0.2V. Thus the voltage on the motor is about 1.3V. The usual motor is rated to turn at 1.5V, so this design is pushing the limits.
Note that the Vf (forward voltage drop) of a diode depends on the current. Usually one assumes a largish current and a voltage drop of 0.6V for a silicon PN diode junction or a voltage drop of 0.2V for a Schottky diode. In this case, the current is so low (1uA or less) that the voltage drop of the diode is less. You can consult the data sheet for the diode to find the Vf versus current curve, extrapolate it (usually), and estimate the actual voltage drop Vf in this circuit.
You might need to choose a different combination of voltage monitor threshold Vth and diode voltage drop Vf. For example, you might try a 1N4841 silicon diode with a Vf that might be around 0.2V at such a low current.
The size (capacitance) of the storage capacitor and the system voltage (say 1.3V) partially determines how much the motor turns when it does turn. (The timing circuit also factors in.) You can use a larger storage capacitor to make the motor turn more revolutions, but less frequently. Too small a capacitor will fail to start the motor. You can use such a small storage capacitor that the motor will only turn a part revolution per turning event, i.e. barely twitches, but does so more often.
This design uses photodiodes for solar cells since they are all that is commonly available in such a small size. Their generated voltage is small (0.35V Voc per cell.) The total generated voltage is about 1.4V. So you don’t have much wiggle room on designing the timing circuit. And the system only works in bright light.
Solar cells generate the most power at maximum power point (MPP.) If the design doesn’t discharge the storage capacitor completely, the solar cells will be operating nearer the MPP for longer. The MPP is typically 80% of the Voc (open circuit voltage) of a solar cell. In this case, the MPP is about 1.1V (80% of 1.4V Voc). So if you trigger at 1.3V and let the system voltage fall only to say 0.8V, you would be operating the solar cell nearer the MPP and the motor would turn more for the same light conditions.
The Voc (voltage open circuit) of a solar cell is the most voltage the solar cell would ever produce (for a given light condition), when it is feeding an “open” or in this case a fully charged storage capacitor. At that voltage, it is producing little power (because there is little current.)
05/28/2019 at 20:52 •
As I said elsewhere, the basic circuit design is ancient, not my work. Search for solar powered bugs using vibrating pager motors. Many use "analog" circuit designs using individual transistors. This one uses IC's and a MOSFET.
A solar panel (many cells, on the sides of the cube) charges a storage capacitor Cstor. A voltage monitor VMon senses system voltage. When there is enough light and the voltage reaches a threshold voltage, the voltage monitor IC turns on an NMOS mosfet switch which switches ground (low side) to a motor. The motor turns a little, discharging the Cstor capacitor, and the system repeats.
The voltage monitor (also called a voltage supervisor or PMIC) has a hysteresis of only 50 mV, which means it turns off again after the sensed voltage Vsense drops 50mV. But the diode Dhold and capacitor Chold form a clever circuit that temporarily holds up the voltage to the sensing pin of the voltage monitor. That increases the time that the voltage monitor output is enabled (low) and effectively increases the overall hysteresis (difference between the system voltage at turn on and the system voltage at turn off.) In other words, the voltage the voltage monitor sees is different than the system voltage Vcc to the motor.
The circuit does not try to store energy through dark periods. There is no blocking diode on the solar panel so the solar panel leaks backwards in the dark. The storage capacitor leaks on the order of one uA. The voltage monitor uses on the order of one uA. The solar cells produce on the order of tens of uA or more.
The Texas Instrument TPS3839 voltage monitor is actually ultra low power, on the order of 150nA. It senses voltage periodically instead of continuously, to save power. But that is moot in this design, you can use other voltage monitors consuming more power. You might want to use another voltage monitor because they are available with a fixed threshold Vth in a limited range. At one time I used a Richtek brand voltage monitor with a Vth of 1.2V and an operating current of a few uA.
See my other blog about tuning the circuit.
05/28/2019 at 20:47 •
The prototype (seen in photos) uses a rigid PC and requires much handwork and soldering.
The main goal of this project is to study whether a flex PCB will bend into a structural, rounded cube. That would save manual fabrication of a cube from rigid PCB.
The bendable parts are absent copper, except for traces that need to cross. The bendable parts only have traces on one side so that they can bend more. Ideally the traces would be on the inside of the bend so they are in compression. In tension, they might break open more easily. But I decided not to use vias to get to the inside of the bend.
The bend radius is 2mm, which is not aggressive.
The other parts of the design (sides of the cube) have copper zones on both sides so that they might be stiffer.
Typically, I fill complete ground zones on both sides anyway, to minimize noise and to reduce copper waste in etching. Here, some of the zones are “no net” because I didn’t want to run traces across the bendable part to ground all the zones.
The sides have SMD solar cells. After bending, the sides will still have bending moments. Solder joints and components are not supposed to carry structural loads. The bending moment would tend to tear open the solder joints to the solar cells. Maybe solar cells oriented the other direction would tear open less easily.
The design has three tabs that fit into slots. The minimum slot width that the OSHPark fab supports is 20 mil. This is greater than the about 7 mil thickness of the flex board itself. Thus the slots are not ideal for structurally gripping the inserted tabs. I considered just drilling the ends of the slots at the minimum drill of 10 mil, and using an Xacto knife to cut a slot between them.