BeagleBone Power Management
Work in progress: plan is to first collect the raw info, then organize and add prose.
|BBone White (TPS65217B)||BBone Black (TPS65217C)|
|LDO1||1.8 V||100 mA||vrtc||LDO1||1.8 V||100 mA||vrtc|
|DCDC1||1.8 V||1200 mA||ddr2||DCDC1||1.35 V||1200 mA||ddr3l|
|1v8||LDO3||1.8 V||400 mA||1v8|
|LDO2||3.3 V||100 mA||power led||LDO2||3.3 V||100 mA||power led (3v3aux)|
|LDO3||3.3 V||200 mA||3v3a||LDO3||3.3 V||400 mA||3v3a|
|LDO4||3.3 V||200 mA||3v3b||U4||3.3 V||500 mA||3v3b|
|U8||3.3 V||1000 mA||3v3exp||3v3exp|
|DCDC2||1.285 V||1200 mA||vdd_mpu||DCDC2||1.325 V||1200 mA||vdd_mpu|
|DCDC3||1.125 V||1200 mA||vdd_core||DCDC3||1.125 V||1200 mA||vdd_core|
Note 1: The TPS65217C configures DCDC1 at 1.5V by default (for DDR3). For the BBB, this is changed to 1.35V (for DDR3L) by software before RAM is released from reset.
Note 2: This shows DCDC2/3 voltage levels after software reconfiguration for maximum performance. Initial voltage levels are 1.1V for both.
Considerations for external connections
Most pins on the expansion headers are directly connected to processor I/O pins, which are supplied by the 3v3a (for drivers, pull-ups, and protection diodes). While power-on-reset is asserted (PGOOD low) these are never driven by the processor but have a (weak) pull-up/down enabled (default pull-direction varies per pin). External hardware must refrain from driving these pins unless the 3v3a has stable power. Unfortunately, neither the 3v3a nor PGOOD is available on the expansion headers, however the following relationships can be used:
- if 3v3a is not powered, PGOOD will be low;
- if PGOOD is low, nRESET will be low (since BBB rev A6) or pulled to 3v3a (before)
- if 3v3a is powered, 3v3exp will also be powered.
Therefore external hardware can abide by the rules by
- accepting the possibility that pins may be driven by the processor when 3v3exp is powered, and
- refraining from driving processor pins when 3v3exp is unpowered or nRESET is low.
These criteria are met for example by powering logic from 3v3exp but making sure its drivers are disabled while nRESET is low.
Some pins have special considerations:
- The power-button signal (P9.09) is weakly pulled up to a PMIC-internal supply. It is intended to be connected to a power button on a physical user interface (if any). It has the rather unique property that its voltage can exceed that of all other pins on the expansion headers (e.g. when in OFF-mode while powered via USB).
- The ADC supply (P9.32) is 1.8V and the ADC inputs (P9.33, P9.35-40) must never exceed it.
- NOTE: be sure to read advisory 1.0.32 of the AM335x errata carefully before making any low-impedance connections to these pins! Oddly, although it seems to me the erratum could cause vdd_adc to be shorted to ground, it claims no workaround is needed (and none is implemented on the BBones) when vdd=vrefp, vss=vrefn, and only moderate-impedance connections are made to the inputs (e.g. a resistive touchscreen).
- The LCD data pins (P8.31-46) are sampled at power-on reset (i.e. when PGOOD goes high) to determine the boot configuration (sysboot 0-15). Unlike other processor I/Os, their internal pull-up/down is disabled by default, and external 100K resistors to 3v3a or ground are used to ensure correct configuration:
- 01 00 0000 00 0 10111 -- BBW (boot order: mmc 0 (μSD), spi, uart, usb)
- 01 00 0000 00 1 11100 -- BBB, normal (boot order: mmc 1 (eMMC), mmc 0 (μSD), uart, usb)
- 01 00 0000 00 1 11000 -- BBB, sd-boot (boot order: spi, mmc 0 (μSD), usb, uart)
- The SD-boot button (S2) of the BBB overrides sysboot 2 by pulling it to ground via 100Ω. External hardware must be careful not to disturb these lines during power-on (unless changing the boot configuration is the intended goal of course). They are only sampled at power-on reset (including exit from RTC-only sleep), not at warm reset.
- On the BBB, the eMMC pins (P8.03-06 and P8.20-25) have 10K pull resistors to 3v3b.
Hardware is a BBB rev C patched to have a unified 3v3 rail, no external connections other than power, probes, and console cable.
Two kernels are compared: the one on the right is a standard kernel + device tree which enables the (reset-insensitive) HDMI framer, while the one on the left leaves the HDMI framer in standby (its default state, negligible power consumption). There are other differences, but those should be irrelevant during shutdown, certainly once power-on reset is asserted (PGOOD low).
Most current consumption is eliminated at the start of the powerdown sequence when PGOOD goes low, although there's a brief burst shortly afterwards (no idea why). Current ramps down more 2 ms later when the core and mpu supplies are disabled, and drops to negligible once the 3v3 supply is cut. Current temporarily increases again due to VDDS leakage until it too is cut.
When the HDMI framer is enabled, its continued current consumption (until 1v8 is cut) is clearly visible.
Same thing, but with the power path being cut by the PMIC at the start of the powerdown sequence, the BeagleBone is now running off its capacitors. In this case it managed to make it without losing regulation at any point, but only barely so. With the HDMI framer enabled the 3v3 and 1v8 supplies receive a premature dent.
The PMIC tries, it really does. In fact, while the 3.3V supplies can't be saved (obviously, with the SYS capacitors charged at a mere 3.4V or so), when not burdened by the HDMI framer it still has 2.4V on SYS by the time the 3v3 is disabled, so it almost certainly would have maintained regulation on the remaining supplies if the VDDS leakage hadn't ruined it.
By the time the PMIC intervenes the 3v3 supply has already lost regulation, and it clearly makes no attempt to sequence the remaining supplies but disables them simultaneously. The HDMI kernel actually initially has a gentler slope since that kernel happens to have better power management. The inexplicable sudden steepening of slope that happens there is worrying.