I am taking a PWM input (0-12 V) from a TLC59711 (driven by SPI from a Raspberry Pi) and I want to drive a P-channel MOSFET.
It is driving a 12 V, 40 W LED. I want to switch at high frequencies, as I want the LED not to flicker with any grade of camera.
High definition cameras can shoot up to 1000 fps, so say 5 kHz (or even 10 kHz). I have chosen this gate driver and a p-channel MOSFET. Can anyone see any obvious problems?
I am new to Spice too - just learning.
Your design looks pretty good.
The 10kΩ resistor is a good touch for pulling the gate high during startup.
I'd swap the 0.1uF out for a 0603 35V 4.7uF or 10uF ceramic cap. That'll have nice and low inductance, plenty of effective capacitance at the 12V bias, and the cost bump is miniscule. (And honestly, there's rarely a reason to go with 0.1uF for decoupling anything low voltage these days, unless you're penny pinching for large scale production - large capacitance MLCCs in 0402 and 0603 have become absurdly cheap and abundant.)
Follow good placement practices for the decoupling cap on your PCB - you want as short a loop as possible between the capacitor and the TC4429's power pins. Keep that PDN inductance low!
An electrolytic bulk decoupling cap (220uF 50V, or 330uF 35V) would be a good idea, too, if the leads to your power supply aren't short. With 12V worth of DC bias derating they should give you enough capacitance to maintain a <1V droop for at least 100us, which should be more than enough to compensate for some stray inductance in the power leads. The ripple current is low enough that you shouldn't have to opt for any fancy low-ESR/DF caps here.
Your MOSFET has a nice low gate capacitance and dynamic parameters, so you should be able to get up to around 10kHz 10-bit PWM at the upper limits. 8-bit will be fine. Remember that the shortest pulse length is the period at the frequency divided by \$2^n\$ for n-bit PWM, so at 10kHz you're actually in the ~400ns pulse range with 8-bit, and the ~100ns range with 10-bit.
The only other thing I'd be doing here is adding a 6Ω resistor inline with the gate, and a 1.5Ω resistor and Schottky diode in parallel with it, for asymmetric drive current, like so:
simulate this circuit – Schematic created using CircuitLab
This allows higher gate drive current during switch-off, which helps eliminate the Miller parasitic self turn-on effect. The gate resistance will also moderate gate ringing. 6Ω is the value indicated in the datasheet for the dynamic timings, at a \$V_{gs}\$ of 10V, so it should get you just as good \$T_r\$ / \$T_f\$ with a 12V drive. You may get away without R2 and the Schottky, because it's typically only important in push-pull or H-bridge designs, where you might get shoot-through, but you'll definitely want R1 in series with the gate.
Make sure you design the board layout as compact as possible. Parasitic inductance is really the killer for these kinds of circuits. The path from the TC4429 output to the MOSFET input should be as short as possible. You also really want a good solid ground plane under your traces to provide a low impedance return current path.
Probably goes without saying, but you'll really want to be making a custom PCB for this. Stripboard doesn't give you a reference plane under your traces, TH parts have tons of stray inductance, and it's hard to route tightly on them. Breadboard is right out the window due to the current anyway, but it's no good even for a lower current prototype because breadboards are full of parasitic inductance and capacitance.
You could maybe simplify this design a little with low-side switching and an N-channel MOSFET, but you'd still need a proper gate driver anyway to achieve the necessary switching speed, so it'd be pretty similar. Since you've already picked out a good MOSFET and done the majority of the design work, I'd say it's not really worth it to redo everything.
Addendum: Having slept on it, I had another thought for a potential design improvement.
Your TC4429's decoupling cap is on the same rail as the load, so when there's a switch-on transient the load will try to pull from all local capacitance. If your layout is tight, the MLCC will be one of the lowest impedance sources of energy. It'll probably be fine, but what you could do is add a Schottky diode in series with the 12V rail that feeds the TC4429. This prevents back-feeding the load with the decoupling cap, so all the transient energy is pulled from bulk decoupling and the power source. Since a very short transient droop on the load side voltage isn't critical, but instability on the TC4429's supply might have greater consequences, this could help improve reliability. You'll probably get on fine without it, but it's worth testing.
There is no reason to actually switch the LED at all. The PWM output, switched via a half-bridge (one PMOS, one NMOS), drives an inductor that converts the PWM voltage into a small amplitude sawtooth current. That can be further filtered with a capacitor. So the LED sees constant current and doesn’t flicker, and you’re not dissipating much power either.
The frequency of the PWM affects the inductance and the capacitance: the lower the frequency, the bigger these two get. So it’s usually a trade-off between size and dissipation. Higher frequency has higher switching losses and potentially higher dissipation in the capactor, but the components get physically smaller the higher the frequency - up to a point where the output power drives the component sizing due to conductive losses and such. At 10kHz you’re far, far away from such limits and in fact you could make the thing smaller by going up in frequency to, say 50-100kHz.
Most generally speaking, any motor bridge driver chip with suitable current rating will do this job. They are designed to drive inductive loads - in your case, the smoothing inductor.
The only reason we have PWM LED flicker is because of cost cutting. To get rid of flicker, add a complementary MOS device or a free-wheeling diode, an inductor, and a capacitor. The full-amplitude PWM current tends to be an EMC concern, and the cables for some even low power LED lights radiate like crazy.
I made an LTspice simulation for a similar application for another recent question. The current limiter is not really needed, but may be useful.
In order to maximize efficiency of an LED drive circuit, an inductor can be used. In this circuit, the MOSFET is driven by a PWM signal. When the device is turned on, current will flow through the LED and the inductor, and when the MOSFET is turned off, the current in the inductor will flow through the flyback diode D2. This process will continue and overall current will increase until the energy added to the inductor during the ON time matches that which is expended during the OFF time of the PWM signal.
This is continuous conduction mode (CCM), and the duty cycle becomes critical as average current is increased. I arrived at the values I used in the simulation by trial and error. Other values could be chosen where duty cycle is not so critical, but at the expense of increased ripple. Eventually it will be running in discontinuous conduction mode (DCM), where all energy stored in the inductor will be dissipated during the OFF time. But that results in very high ripple.
