TPS2590RSAR
Product Specifications
TPS2590RSAR
Texas Instruments — Power Management & Signal Chain
Product Overview
The TPS2590RSAR from Texas Instruments is a power management / signal chain IC delivering high efficiency, precision regulation, and robust protection. Texas Instruments (TI) is a global leader in analog ICs, power management solutions, and embedded processors.
Superb Automation partners with leading brands including Qorvo, TI, ADI, XILINX, NVIDIA, Intel, ST, MACOM, Renesas, and Coilcraft to provide complete BOM and supply-chain solutions.
Key Features
- High-efficiency power conversion with low quiescent current
- Wide input voltage range with precision output regulation
- Integrated over-current, over-temperature, and short-circuit protection
- Small footprint packaging for space-constrained designs
- AEC-Q100 qualified options available for automotive applications
- Flexible sequencing and soft-start control
Key Specifications
| Input Voltage | V |
| Output Voltage | V |
| Output Current | A |
| Switching Frequency | kHz |
| Efficiency | % |
| Quiescent Current | µA |
| Topology | — |
| Package | — |
| Operating Temp | °C |
| Protection | — |
Applications
Request Quote — In Stock, 100% New & Original, Full Warranty
TPS2590RSAR: a 5 A, 20 V load switch that carries two separate current thresholds, a programmable fault timer and a choice of latch or retry after a fault, so a line card can ride through a disk spin-up and still be protected against a short
TPS2590RSAR is an integrated hot swap switch with the pass FET on the die, a fault current threshold that starts a timer, a higher hard current limit that acts at once, and an internal power limit that keeps the FET inside its safe operating area while a large output capacitance charges in a QFN body of 16 pins. The RSA body is sixteen pins on a 4.00 mm by 4.00 mm pad with an exposed thermal pad underneath, and the part exists because a fuse cannot tell a disk drive spinning up from a shorted backplane.
Order code and package
| Catalogue string | TPS2590RSAR, where RSA is the sixteen pin QFN body and R is the large tape and reel |
| Body | QFN, RSA, sixteen pins, 4.00 mm by 4.00 mm, exposed pad on the underside for heat into the plane |
| Reel | 3000 pieces on a large tape and reel, the same count on the plain row and on the G4 row |
| Finish and moisture | NIPDAU plating, moisture level 2 at 260 C, one year of floor life once the bag is opened |
| Temperature | minus 40 to 125 C junction, which is the figure the data sheet quotes rather than an ambient range |
| Marking | TPS on the first line and 2590 on the second |
| Document | SLUS960G, July 2009, revised January 2014 |
| Status | Active and in production on the addendum dated November 2025 |
Electrical parameters that decide the application
| Bus voltage | 3 V to 20 V of operating bus, with 25 V absolute maximum on IN and OUT, so a 12 V or a 19 V rail sits inside the rating and a 24 V rail does not |
| On resistance | 29.5 mOhm typical and 42 mOhm maximum measured from 1 A to 4.5 A, which is the whole conduction loss budget at 3 A |
| Fault threshold | IFAULT is set by RIFLT and prints 0.8 to 1.2 A at 200 kOhm and 1.8 to 2.2 A at 49.9 kOhm, and the vendor asks for it never to be set below 1 A |
| Hard current limit | ILIM is set by RILIM and prints 1.6 to 2.4 A at 100 kOhm, 2.6 to 3.4 A at 66.5 kOhm and 4.6 to 5.4 A at 40.2 kOhm |
| Fault timer | a 35 uA current charges the CT capacitor towards 1.4 V, so the timer is 40000 times the capacitance, which is 4 ms at 0.1 uF and 40 ms at 1 uF |
| Retry against latch | RTRY above 2 V latches the switch off until EN is cycled, RTRY below 0.8 V retries once the capacitor has discharged to 0.16 V at 1.4 uA |
| Enable | active low, with a 0.8 to 1.5 V threshold and 50 to 250 mV of hysteresis, so a slow ramp does not chatter the switch |
| Switching speed | 350 to 500 us from EN falling to 90 per cent of the bus and 10 to 20 us to turn off, which is the number that decides whether the rail collapses on a fault |
| Supply current | 25 to 100 uA with the switch off and 3.9 to 5 mA with it on, because the internal charge pump runs while the FET conducts |
| Reverse conduction | the body diode drops 0.77 to 1 V when the output is held above the input, so a back driven bus is not isolated |
| Thermal shutdown | 160 C with 10 C of hysteresis, and the fault flag is a separate open drain output |
| Approvals | recognised to UL 2367 under file E169910, which matters for a plug-in board that has to pass a safety review |
Reading the two thresholds against a real load
| A 3.5 A board on a 12 V bus | RILIM of 40.2 kOhm gives a hard limit of 4.6 to 5.4 A, above the steady current and inside the 5 A ceiling of the part |
| The same board with RIFLT at 49.9 kOhm | the timer starts at 1.8 to 2.2 A, so an overload that lasts longer than the timer removes the load without reaching the hard limit |
| CT at 0.1 uF | about 4 ms of ride through, enough for a 1000 uF output capacitance charged at 3 A and not enough for a 4700 uF bank |
| CT at 1 uF | about 40 ms, which covers a large bank and also means a real short is left on the bus for 40 ms before the switch opens |
| RTRY tied to ground | retry behaviour, the switch re-enables after the capacitor falls to 0.16 V and keeps trying until the fault clears |
| RTRY tied to 3 V | latch behaviour, and the board stays down until somebody cycles EN or power, which is what a telecom line card usually wants |
Points to check before this part goes on a BOM
The two thresholds exist because one threshold cannot do both jobs. A disk drive pulls several amps for a few hundred milliseconds when it spins up, and a single current limit set just above that figure leaves the board unprotected against a partial short that draws the same current for minutes. Here the lower threshold, IFAULT, only starts a timer. If the current falls back under it before the timer expires, nothing happens. The higher threshold, ILIM, acts at once and is set above any legitimate surge. A board that sets both resistors from the same table row gets the worst of both worlds, which is why the two are printed separately.
The fault timer is the part most often set wrong on a first article. The capacitor is charged with 35 uA towards 1.4 V, so the timer follows 40000 times the capacitance and nothing else. A 0.1 uF capacitor gives 4 ms. If the output bank is 1000 uF and the load pulls 5 A, the bank needs 2.4 ms to reach 12 V and the timer has 1.6 ms of margin. Change the bank to 4700 uF and the same board needs 11 ms, the timer expires mid ramp, and the switch opens on every power up. The symptom looks like a dead board and the cause is one capacitor value.
Power limiting is the second protection layer and it is not a current limit. While the output rises into a heavy capacitance the device holds the FET inside its safe operating area by dropping the current as the drain to source voltage falls, and the fault timer runs throughout. That is why the timer has to be longer than the charge time rather than longer than the load surge. A board with a very large bank and a short timer will fold back and shut off, and the only honest fix is a longer timer or an inrush resistor, not a bigger current limit.
Latch against retry is a system decision, not an electrical one. In retry mode the switch re-enables after the CT capacitor falls to 0.16 V and the fault is still present, so the board sits in a hiccup loop and the bus sees a short every few tens of milliseconds. On a telecom line card that is usually unacceptable, because one failed card can disturb the backplane. In latched mode the switch stays off until EN is cycled, the fault flag stays asserted, and the system controller can decide whether to retry. The RTRY pin is read as a level, so a resistor divider or a controller output both work.
Reverse conduction is the gap in the protection. When the output is held above the input, the body diode conducts and drops 0.77 to 1 V, so a back driven output feeds the bus through the switch. On a board where a downstream converter can hold the rail up after the bus is removed, that path is real and the diode drop becomes heat in the package. The fix is a second switch in series or an external diode, and the data sheet does not pretend otherwise.
The enable pin is active low, which catches people out. Pulling EN to ground turns the switch on, and letting it float or driving it high turns the switch off after the 350 to 500 us propagation delay. A board that ties EN to a controller with an open drain output has to check the pull up, because 350 us of delay on a 3.3 V rail is enough for a processor to start booting from a rail that is about to disappear.
Setting the pair of resistors from the load, not from the table
Start from the steady state current and add the tolerance of the load, not the tolerance of the resistor. A board that draws 3 A steady and 4.2 A during a spin-up needs a hard limit above 4.2 A plus margin, so the 40.2 kOhm row at 4.6 to 5.4 A is the only printed option that fits, and the 66.5 kOhm row at 2.6 to 3.4 A will trip on the surge. Then set IFAULT below the steady current, not above it, because the point of the lower threshold is to catch a load that has degraded into drawing more than it should for a long time.
The resistor tolerance is not the only spread. The thresholds print a plus or minus 10 per cent window at 25 C and that window is what the last column of the table describes. Two boards built from the same 49.9 kOhm resistor can trip at 1.8 A and 2.2 A, so a load that sits at 2.0 A steady is inside the window of both and will trip on some units and not others. Move the steady current down to 1.5 A and the ambiguity disappears.
What the flag output does and does not tell you
FLT is an open drain output that pulls low for an overload that timed out or for a thermal shutdown, and it becomes operational before the undervoltage lockout releases, when the input is above about 1 V. It also pulses low briefly during the ramp of the output on power up, which a controller that logs the flag as a fault will read as one fault per power cycle. The pulse is short and the workaround is a blanking window in the firmware or an RC filter on the flag line.
The flag does not distinguish a current fault from a thermal fault, and it does not report the power limit mode. A board that needs that distinction has to read the CT pin or watch the current, which is why the flag is best treated as an interrupt rather than as a diagnosis.
Where this order code fits
This is the switch between a backplane and a plug-in card on a telecom shelf, a RAID enclosure, an SSD carrier or a PCIe card, where a board has to be inserted into a live bus and a short on that board must not take the shelf down Teams comparing it with a neighbouring part start from our Power Management Cross Reference page, and teams replacing a part that has gone end of life start from the Component Obsolescence Management page.
Questions buyers ask
Does it need an external FET? No. The pass FET is on the die and the 29.5 mOhm typical on resistance is the switch itself. An external FET would only be needed for a current above the 5 A rating.
Can I set the fault threshold to 0.5 A? The vendor asks that IFAULT is never set below 1 A because the threshold accuracy is only specified above that point. A load below 1 A needs a different part.
How long can the output capacitor take to charge? As long as the fault timer allows. With 1 uF on CT the timer is about 40 ms, so a 4700 uF bank charged at 3 A fits and a 22000 uF bank does not.
What happens on a short at the output? The current rises past ILIM and the switch opens within 10 to 20 us without waiting for the timer, then retries or latches according to RTRY.
Does it block reverse current? No. The body diode conducts from output to input and drops 0.77 to 1 V, so a back driven output is not isolated from the bus.
SUPERB holds and sources this line for production and repair demand, quoted against the exact ordering string with date codes, packing and country of origin stated on the offer. When a part is moving, the How to Verify a Drop in Replacement page sets out the checks that keep a substitute from failing on the bench.