Soft Start Circuit Design: Ramp, Inrush and the MOS Between Them
A soft start circuit is usually described as a capacitor that slows the gate, and usually implemented by copying a resistor and capacitor value from another design. On the bench the output ramp looks smooth, the board passes, and the assumption survives until a different load or a different input cable produces a failure that nobody connects to the start-up ramp.
The reason is that two distinct jobs get folded into one number. One job is to let the input settle before the main path conducts, and the other is to control how much current the downstream capacitance is allowed to draw. The two can share hardware, but they cannot be reviewed with the same question.
Contact Bounce and Capacitive Inrush Are Different Problems
When a connector is mated by hand or a mechanical switch closes, the contacts may make and break several times in a few milliseconds. During that interval the input voltage is not a step, it is a burst of steps, and logic that enables the main path on the first rising edge can be enabled, disabled and enabled again before the input is actually stable. The first requirement is therefore a delay that survives re-triggering, measured from the last bounce rather than the first.
At the same time, the downstream storage capacitance presents what looks like a short circuit at the moment of connection. The inrush current is set by the input voltage, the source impedance and the total capacitance, and it can be far above the steady state current of the load. A diode or a fuse chosen on the steady state number will be stressed on every connection, and the connection is the moment when the current is least visible on a bench supply with a current limit.
A Voltage Ramp Is Not a Current Limit
Slope control works by holding the gate in the transition region so the output voltage rises slowly. The output current, however, is still determined by the load impedance, the downstream capacitance and by when the load decides to start drawing current. The same voltage ramp applied to a light load and to a heavy one produces two different current profiles.
If the requirement is a defined current ceiling, the implementation has to sense current and act on it. A circuit that only shapes the gate voltage cannot promise a limit it never measures, and an output that looks smooth on one channel of the scope says nothing about the current in another. Confirming the limit is a measurement task, not a waveform review. Reviewing the ramp and reviewing the current limit are separate exercises, and the second one needs a current probe or a sense resistor, not a voltage waveform.
The MOS Is Not a Switch During Start-Up
While the gate voltage is between the threshold and the fully enhanced value, the device is a resistor with a large and rapidly changing value. It simultaneously carries load current and sustains a drain to source voltage, so it dissipates power at a rate that can be orders of magnitude above its steady state loss. How long that interval lasts, how much current flows and how well heat can leave the package are all variables, and all of them belong in a safe operating area check against the specific part. Thermal design across the rest of the board sets the starting condition for that check, so the pass device cannot be reviewed in isolation from its layout.
The gate to source voltage deserves its own line in the review. A long ramp usually comes from a large gate capacitance or a large series resistance, and neither reduces the voltage the gate oxide sees. Startup count also matters: a connector that is mated many times over the life of the product repeats the transition thousands of times, and a device that survives one event has not been characterised for that many.
Why the Load Restarts at Half Voltage
Downstream converters, processors and supervisory circuits have their own turn-on and undervoltage thresholds. If the supply ramps too slowly, the load can start drawing current while the rail is still below its minimum, which pulls the rail down, which resets the load, which removes the current, which lets the rail rise again. The symptom is a supply that starts, fails, and starts again in a slow oscillation that looks like instability in the regulator.
Three quantities settle the question. The first is the load turn-on threshold and the current it needs at that threshold. The second is the power sequencing requirement when more than one rail is present, including which rail must be valid before another is allowed to rise. The third is the discharge path: after a fast power down, the soft start capacitance has to be empty before the next connection, or the next start begins from an intermediate state instead of from zero.
Reviewing a Soft Start by Measurement
State the targets for debounce and for inrush separately, and do not let a single time constant stand in for both. Record the input voltage, the output voltage, the gate waveform and the load current together, so the phases of the start-up are identifiable in one capture. Repeat the capture with no load, with a typical load and with the worst case starting load, then check the transition stress, the gate to source rating, the temperature rise and the effect of repeated mating. Finish by exercising slow power down, fast power down and immediate reconnection, and confirm the reset behaviour each time.
That last test is the one most often skipped and the one most often responsible for a field return, because a supply is rarely connected once, and in a product the connector is a purchased part whose contact behaviour varies between suppliers. The behaviour of the whole board under repeated connection is also something the assembly and test stage can verify on a sample basis, which turns a bench observation into a production criterion rather than a personal habit.
Three Beliefs That Cause Trouble
Longer is not automatically safer, because a slower ramp extends the transition region, raises the energy dissipated in the pass device and can leave the load parked below its undervoltage threshold. A smooth output is not sufficient evidence, because the interesting quantities are on the current and gate channels. And a copied resistor and capacitor pair does not transfer, because input range, load capacitance, device characteristics and the start-up policy of the load all change the result.
FAQ
Is soft start the same as inrush limiting? They overlap, but a ramp that limits voltage rate does not by itself guarantee a current ceiling.
How long should the ramp be? Long enough for the input to settle, short enough to keep the pass device inside its safe operating area and the load above its threshold.
Can one circuit do both jobs? Often yes, provided each requirement is verified with its own measurement rather than assumed from the other.
What is the most common oversight? Testing with a single load and a single connection, which hides both the restart loop and the repeated stress.
Summary
Designing a soft start circuit properly means separating three questions: does the input reach a stable state before the main path conducts, does the charging current stay within the limit the components were chosen for, and does the pass device stay inside its safe operating area for the whole transition. Answer them with simultaneous measurements of voltage, gate drive and current, on a warm board and a cold one, with the load the product will actually see. A supply that survives a single clean connection has only proved that the simplest case works. The records from those tests are what make the behaviour repeatable across builds.
