This circuit uses a negative buck regulator to regulate -5V from a -12V input. The -5V output is only allowed to rise after the system main +5V has come up, and it automatically shuts down the -5V when the +5V goes down.
The circuit in Figure 1 steps down a nominal -12V to a regulated -5V. It allows -5V to come up only after a separately regulated +5V has come up, and if the +5V collapses, it automatically shuts down the -5V. This is useful in ±5V supplies for A/D and D/A converters, which often require such power-supply sequencing to avoid latchup.
IC1 is a conventional boost regulator, but the overall circuit is a negative buck regulator. The boost-regulator topology is correct for the switching control, but the regulator's feedback signal—which monitors an output voltage referred to the converter's positive rail and compares it with a reference voltage referred to the converter's negative rail—requires a level shift. The Q3/Q4 current mirror provides this shift, with emitter resistors R8 and R9 included to minimize the Vbe-mismatch error.
IC1 includes a comparator and a 1.5V reference, normally used for low-battery detection via LBI and LBO, which monitors the +5V rail as follows: the current in Q1, mirrored by Q2, flows through R4 and develops a voltage proportional to the +5V rail. If this rail falls below a nominal 4.2V, the LBO output pulls R5 to the negative rail. That connection causes a current increase in the diode-connected Q4 which, mirrored by Q3 and flowing in R3, causes a rise in FB voltage to the regulator.
Feedback as described above tells the regulator that no additional output energy is required, so it complies with a shutdown in which the internal pulse-frequency modulation (PFM) suspends all power-conversion cycles. Connecting a minimum load of 10kΩ will prevent leakage through D1 from charging up the output capacitor (C2) while in this state. When IC1 operates with a +5V input and as a boost converter (as intended), it delivers about 150mA from a +12V output. The buck-regulator configuration, on the other hand, delivers 400mA at -5V using similar high-current components.
Efficiency vs. load current measures 85% at 100mA, 89% at 250mA, and 90% at 400mA. The measured peak-to-peak ripple is less than 25mV for any load. Output-voltage accuracy depends on the 2%-accurate reference in IC1 and the tolerance of feedback-path resistors R1, R3, R8, and R9.
Any difference in Vbe for transistors Q3 and Q4 introduces an additional error. Vbe measures about 550mV for the transistors used, and the maximum Vbe difference measured among Q1–Q4 was 9mV. With respect to the Q3–Q4 base voltage (-1.24V), this 9mV contributes another 0.75% error in the output voltage. To match the Vbe drops to within 1mV and eliminate the R6–R9 resistors, substitute a dual transistor such as the Rohm UMT1N (available in a SOT23-6 package).
A similar idea appeared in the 9/98 issue of Electronics World & Wireless World (UK).