BootSwitch - Best Practice in Bootmanagement

Top | Index

Hardware details

General Note: In this document, i will try to coherently use the term "standby" to refer to the technical standby state, which is equal to a "switched-off" computer, whoose power supply still delivers the standby voltage. (ACPI mode "S5"), while "Power-On" denotes the powered-on PC, where all regular voltages are provided and booting can of an operating system is imminent (ACPI mode "S0").

Basic Principles: This boot drive switch is to connect the chosen hard drive to the PC system, even before the PC is powered-on again. What we get is the most gentle, most compatible procedure of changing a boot disk, since the inevitable power upsurge can not be harder as if the drive was permanently connected to the system, thus being supposedly within all standard limits. Thus, for the hard drive, power supply and BIOS the electrical and logic situation is exactly the same as with a "hard-wired" system, except from the fact that the drive may have been a different one at the time of an earlier power cycle.
Now you may ask: Where does the electronics gets its power to interprete button events and energize some relay when the PC is not powered-up at this time? Extra power supply? Batteries being recharged at runtime?
Even more simple! When an ATX power supply is turned "OFF", all the regular power rails are in fact switched off, except from a 5 volts of standby voltage. This 5 volts will remain available as long as the PSU is still being connected to a live mains.
The 5VSB will usually feed some circuits that may be utilized to "wake-up" the system from standby state. These are namely the circuit for the power button, but also certain timers, keyboards, network cards, and possibly some other stuff.
The 5VSB is often generated by a separate converter in the power supply, so its efficiency should be acceptable. According to the ATX specifications, the 5VSB are resilient to at least 1 Amps.
So, the power supply of a small circuit add-on from 5VSB is no electric problem, but rather a mechanical challenge. Unfortunately the ATX standard does not provide us with a separate connector of 5VSB. Yet we can easily tap the voltage from the ATX mainboard connector. Alternatively the power could be tapped also from a PS/2 or internal USB port. (Refer to your mainboard's documentation!)

Microcontroller: A boot disk switch must be aware of the actual state of the PC system to safely prevent any unappropriate switching at runtime. A clear distinction between "Power" and "Standby" can be made from the existence or non existence of regular board voltages. In the "standby" state, the bootswitch will have to count button events and then engage connection of the chosen hard disk device. However, if the regular voltages are present then the PC is already in "Power-On" state and no further switching operation is to be allowed. Button pushes may be passed-through to the mainboard, so we can make use of ACPI functionality (shutdown or energy saving mode by power button).
It occurs, that this basic functionality could have been implemented with some good old 4xxx or 74xx logic gates as well. Yet considering the "exceptional conditions" (real coldstarts, voltage drops or instability), it would have led to pretty much additional efforts in discrete logic.
On the other hand, we can do all this (and much more) with a small microcontroller of today. I have chosen the ATtiny2313 for this application because of the very good long-term performance experience in other projects (e.g. LED Clock, CodeLock). As with most AVRs, the ATtiny2313 already has built-in features to ensure clean start and operating characteristics.

Switching Relays: Connecting the desired HDD to the board power is done by regular (mono-stable) relays in this concept. Relays are energized before the board voltage rises up. By using relays, we evade from nonlinearity and voltage drop like with semiconductor-based switching methods. The contact set of a good relay behaves just like a good connector or a good mechanical switch.
With the proposed relay type (contact capacity: 8A), the hard switching relays could handle several amps of initial current to even handle a massive 3.5 inch HDD.
Switching relays of that kind are hardly to find with coil voltages of about 5V. Using a type 6-V-relay in a 5 volts appliance is not recommended. Considering additional voltage drop across the switching transistor and admissible undervoltage on the 5-V-rail, the actual voltage to energize the relay may indeed fall slightly below the specified minimum. This could rise critical/unsafe conditions and should be avoided .
We can avoid this by using 12-V-relays. In a switched-on PC, the relay may be permanently hold by the regular 12-V rail. Voltage tolerances are much higher compared to the 5- or 6-V-types. Incidentally we get clear separation between logic and control circuit by use of a 12 V relay. But now another question arises: Where the hell do we get 12 volts from to energize our relay, when there is only 5-V-standby available in a PC? See the next section!

Charge Pump: This special circuit, consisting of IC3 (MAX232), and some peripherals, makes use of the charge pump, inverter, internally generated switching frequency and both of the line-drivers to transform the supply voltage of approx. 5 V to a square-wave voltage of approx. +/-8V (@ 50 kHz). By the voltage doubler (C11, D2, D3) it will charge up the electrolyte capacitor C12 (1000 µF/25V) within few seconds to peak voltage of about +16V. Accumulated energy is pretty sufficient to activate any standard relay in question with coil rating of 12 volts.
The line drivers of the MAX232 are current-limited (short-circuit-proof). On one hand, this makes the circuit virtually "indestructable". However, it could not hold every relay for an unlimited time. After a few seconds the large relay type (Finder series 41) would drop-back at an unforeseen time because of insufficient holding current. It is better if the controller could deliberately initiate the shutdown after some well-defined holding time.
The bypass of 5VSB over D1 makes it happen! After the initial charge of the capacitor was used up, the bypass prevents further voltage drop below approx. 4.7 V minus 0.6 V (voltage drop across the switching transistor), so we are always dealing with a minimum voltage that will be sufficient holding voltage for this relay type.
It is now up to the BootSwitch firmware to fully control relays activation time. If the regular power does not appear, it will trip off the relay after approx. 5 seconds. That's a pretty big time window even for a manual start of the PC (by separate power button). By the way, the timeout provides a convenient functional testing.
D1 to D4 must be Schottky diodes to minimize voltage drop. We've chosen the quite robust SR160 (60V/1A). Especially D1 and D4 are basically oversized, but it won't hurt to have some safety margin at this point, because these diodes separate circuit from PC board voltages.
The charge pump approach has clear "environmental benefits" for the PC system. Other than step-up converters or those miniaturized inverter modules, no further RF interference is generated and in the load case, no current-peak spilling over to 5VSB, since the initial power to energize some relay will be delivered primarily by the charge capacitor. If the PC-PSU was sound, after some tenths of a second, the regular 12 V voltage takes over the entire coil current. Idle current of this MAX232 charge pump is only about some milliamps.
This circuit design has been successfully tested with several specimen of MAX232N (industrial/standard). If ever possible, we should try to pick some chip from original manufacturers (i.e. Maxim or TI). NOT SUITABLE IN THIS DESIGN: MAX232A OR MAX232CPE.
All electrolyte capacitors should be "105-deg"-types sein, that have better durability despite of hot operating conditions. As you can see on the Photo, we can also use unpolarized MKS film capacitors for C7-C11 (1µ/63V). These won't degrade within a realistic period of time.

Relay Drivers: The relay coils are energized by generously dimensioned NPN-switching transistors T1x (BC337-25). Resistor R8x (1k0) injects a base current of approx. 4 mA if the controller rises the respective PORTB to active HIGH. This is an appropriate overdrive factor for this transistor. The relay in the collector branch will respond quickly. For a short time, the coil current is about 50 mA (measured).
At this time, the board voltages are still down, so the "switched" HDD won't experience any current pulse.
Right after some 1/10 seconds, the PSU should come up with the regular 5 - and 12-V voltages for drive supply. Also the already energized relay is now fed by the 12V and coil current adjusts to around 40mA. Now T1x D5x and R9x provide a permanent base current from the 5-V-rail that is now independent from the microcontroller's state. The relay is safely "latched", since it cannot drop back as long as the regular voltages are sufficently stable. Only a complete shutdown (power-off) of the system will drop the relay.
All current values have been dimensioned with generous safety margin. See datasheet to the BC337.

Button-Input:  We may connect any simple pushbutton with normally-open contacs to the terminals X2 "BTN-IN" of the BootSwitch PCB. This button is the sole input device to choose a boot drive by "counting code". After the designated HDD has been connected to the PSU rails (which are yet dead), we would have about 5 seconds of time to manually restart the PC by pressing the regular "Power" Button in this "manual" setup. Otherwise, the relay will drop after this period of time.
Why bother with TWO buttons?! We could simply make the existing Power-Button the input device for the BootSwitch! Simply pull its connector from the mainboard and plug it to X2-BTN-IN. (Power button is often identified by its black-white cable and a connector in 2.5-mm grid is sometimes labelled "PWR".) Now, who's gonna power-up the system via mainboard? This is enabled by the...

Button-Output: After it has energized one HDD relay, the boot switch will provide a compatible Power-On-Signal for the PC mainboard at X3 "BTN-OUT". We may directly connect this output by means of any suitable two-wire cable with the now free connector on the mainboard. The optocoupler IC2 (PC817) delivers the potential free switching signal. This will circumvent possible issues with ground lead and unusual voltage levels. Of course, polarity must be observed, since the output of the coupler is a bipolar phototransistor.
Instead of testing for the right polarity, we should simply try it out. There's nothing that gets damaged with wrong polarity and the chance that we correctly connect the plug on good luck, is something around 50 percent... (According to Murphy the plug at the first attempt is connected basically upside-down. No reason to calm ;-)
For security reasons, the optocoupler IC2 is driven "low-active". The anode of the internal IR-LED is tied to the positive terminal of operating voltage, while its cathode leads to the port output PD5 by way of a current-limiting resistor. To activate the optocoupler, two conditions must be met: The portline PD5 must be configured as an output AND it must be pulled down to a logical Low. Resulting current for the transmit LED is about 15mA at R4 (220 Ohms). This is possible only within the respective subroutines of the programme. It can't happen "by accident".
For instance, when hardware reset was applied to an AVR via /RESET pin, the controller turn all port lines to input and switch off all internal pull-ups (= high impedance) at the instant. No current could be sourced or sinked by the port either against GND or Vcc. Even if the Pull-up on the respective portline was still activated for any reason, no significant LED-current will occur with this circuit. In addition, there are programming measures in place to safely avoid uncontrolled activation of the optocoupler in case of a firmware hookup etc.

Find datasheets for all important components here.

Top | Index

---

Firmware

BootSwitch-Firmware has been written in genuine Assembler, again to achieve the necessary reliability and transparency. By the way, some of the most important features of this application would have been impossible in a so-called "high level" programming.

• Repeated Resets: As mentioned in the hardware description, the microcontroller is being provided a /RESET pulse by every button push event. Should the controller hangup for any reason, it will be re-initialized by the first button event and the application should be workable immediately (provided that the controller and the surrounding circuit are in order). The counting procedure is perfectly working despite of controller resets, because the firmware buffers vital variables to the internal SRAM. Memory contents of SRAM will survive any controller warmstart or Reset. (More precisely: in Assembler programming, we are not obliged to run standard initialization routine that overwrites all SRAM contents.) On the other hand, the functionality of counting button events is not depending on these resets! Button events that may have not triggered some reset for any reason, are also counted correctly, as long as they have a certain minimum length (debouncing). If you don't believe, just remove the coupling capacitor C5!
  

• Reset Blocker: While the firmware generates relay- and power-on-signal, there should be no interrupt from any source. In particular, no hardware Reset should occur, because it would break the corresponding timing loop and the generated pulse could be too short and instable switching could be the result. Firmware needs to block further external reset pulses from the controller. This is possible by reconfiguring the port for button input (PD4) temporarily as an output and setting this port line to a permanent low level. Pressing the button will not produce a high-low transition then, and no negative pulses can be forwarded by C5 to the /RESET input of the controller. This way protected from external Resets, the controller is running its critical timing loops to perform the relay- and power-on timed sequence. After this is done, it will re-enable the push button input.
  Internal Reset sources, like Watchdog-Reset, are not affected by this blockade, see next section.
  

• Watchdog-Reset: Should any critical routine hang up for longer than 8 seconds, the watchdog-timer would be alert and issue an internal Reset to the controller that will re-initialize the firmware.
  

• Other Reset Events: Firmware must only count the "real" button events. Power-On-Reset (coldstart of the controller) or Brown-Out-Reset (undervoltage) are reset events, that will set certain flags in the MCU status register, which allow to differentiate such events from external Resets that have been applied by button presses. This helps to prevent coldstarts and voltage fluctuations from triggering the counter.
  

• X2 (BTN-IN): In "Standby-Mode" (PC Power OFF, only 5VSB existing) the firmware is counting how often the button has been pressed in short sequence. A pause of more than 1 second will be interpreted as the end of submission. The firmware will then block further resets (see above) and perform the required tasks.

• Relay Control: Shortly before activating some relay, the firmware must make sure that there is sufficient charge to safely energize the relay. If this is not the case, the firmware will wait up to several extra seconds until Watchdog-reset occurs. Still in Standby-mode, the User will have another try.
  The voltage level in the load capacitor is being sensed by port input PD2 behind the divider R2/R3. Controller sees a barely valid high level, as soon as the voltage charge has reached 10 V at least.
  To energize some relay, the firmware will issue a logic High level to the respective PORTB line. The associated switching transistor that will then drive the relay's magnetic coil. The relay-on-signal will be held for at least 100 ms, followed by the issue of a PC power-on-signal. If power-on was successfull (regular 12V and 5V appear) the respective relay will be latched by the hardware feedback (see circuit description). However, the firmware will continue to hold the respective PORTB line at output and on High level as long as possible so that the selected transistor receives additional base current from the controller for the runtime of the PC system. This is just to another precaution to avoid dropback of a relay that has been energized if there are some severe problems with the surrounding hardware.
  PORTB outputs that are not supposed to drive a relay transistor stage, are always being set to active Low level, so that they are tied to GND by the low impedance of R8x. This is supposed to further minimize receptability for electromagnetic disturbance.
  

• Output of Power-On signal to X3 (BTN-OUT): After energizing the chosen relay, the firmware is to provide some Power-on-signal for the mainboard that will restart the PC system (PD5). Though technically sound PSU's should provide all regular voltages earlier than within 0.3 seconds after receiving a Power-on request, the firmware generates this power-on signal at least for 0.3 seconds of duration. If the voltages are not present yet (checked from the 5-V-rail), the firmware will continue to hold the signal (equal to holding the Power-button). Should the Power-On have failed after 5 seconds, firmware will stays Standby mode.
  With successfull Power-On, the respective relay control at PORTB is being held on active High level for all the further PC runtime.

• Passing button-pushes in Power-On-State: While PC is running, the firmware must not count further button pushes and not trigger any relay switching. Instead, button-pushes are being forwarded more or less directly to BTN-OUT. This enables to make use of the power-button ACPI functionality. The existence of regular 5V is being checked regularly to make sure that the firmware detects any transition between Power-On and Standby, even if the PC was shutdown by software.
  

• Undervoltage detection: Under the conditions of a real coldstart (PSU been disconnected from the mains), the 5VSB may be instable at the beginning. The Brown-Out-Detector (BOD) in the AVR is to be activated and configuered to a threshold of 4.3V to make sure that the controller will start programme execution only after the supply voltage has reached a safe value above 4.3V (Note: Allowable tolerance on the 5-V-rails is +/- 5% according to ATX standard. The 5V and 5VSB should never drop below 4.75V under regular operaton.)

• Programming of Flash & Fuses: BootSwitch Firmware is to be written via ISP to the flash memory of ATtiny2313 only. No EEPROM data and no specialist things to be observed. Fuses must be set according to the instructions that are given in the Download-Packet.
  

Top | Index

---

Software

BootSwitch does not require any software support.


Top | Index

---

Making of...

Circuit board: My layout was designed for unilateral PCB of about 100 x 75 mm (or 100 x 80 mm). The download provides you with an artwork of 300 and 600 dpi of pixels resolution.
If we plan switch some very power consuming 3,5-inch HDDs, a reinforcement of the 5- and 12-volt paths may be considered. Alternatively, if available, one may choose base material with double copper laminate.
Do not forget to wrap the 2 bridges under Relay Re1-3 and Re1-4 ! They are vital for the microcontroller to detect sufficient Relay-charge and regular board power.

Relays: DIL-type low-profile 12V, 2 changeover, rated 8A (e.g. Finder 41.52.9-12V). For small drives, i.e. Notebook HDDs or SSDs, we may also use the 2-Amps-rated DIL relays (e.g. Finder 30.22.9-12V). My PCB is compatible to both relay pinouts.
Relay contacts: Since the load circuit is being closed even before the regular board voltages rise up, the relay contact's wearout is quite minimized. Yet, there is no indication that the contact set could overduly oxidize or corrode in this application, since from the moment the PSU was reenergized, there will be well enough minimum in the load circuit, i.e. the operating current for the switched HDD. Let's have a look at the datasheet to the finder series 41 relay: The minimum load was specified to about 300 mW. Any conventional hard disk drive will have multiple of this power consumption. Even SSDs need more power these days. (Intel Series 330 SSDs: 600 mW idle, 850 mW access). That is to say: Don't worry about her wetting current.

Power connectors: The power supply to be forwarded to a "switched" HDD is delivered to the BootSwitch PCB via X4, which may be a standard Molex connector (male) for direct soldering.

No detachable connectors are recommended for the switched terminals X5x. Unlike commercial crap, we should not deal with those tiny "floppy" connectors in the 2.54-mm-pitch, since those are quite under-dimensioned with regard to the current consumption of bigger 3.5 inch drives.
Instead, we should customize suitable cable harness with Molex or SATA-power connectors that will be directly soldered to the PCB of the BootSwitch. See photo of the sample board! No additional and possible loose contacts.
Refer to the Pin 1 indication on the standard connectors, the PCB (a small triangle) and the standard colours of the cables!

Drillings:
0,8 mm for most through-hole components; 1,0 mm for connectors X1-X3 and relays;
1,2 - 1,5 mm for the power connectors; 3,0 mm for M3-mountings

Top | Index

---

Howto obtain 5VSB from ATX-PSU

Top | Index

Last revision: 12/2014, 01/2015, 08/2015, 07/2017

---