Shop OBEX P1 Docs P2 Docs Learn Events
P2 hardware reference design and choices - Page 3 — Parallax Forums

P2 hardware reference design and choices

13567

Comments

  • jmgjmg Posts: 15,140
    .... I will look into putting some other useful stuff on the breakouts such as the USB cable etc. All the files are in my public Tachyon folder but here is a link to the P2 folder, just look in PCB and P2PCB etc although I haven't done a final export yet.

    BTW - here are shots of the board as it is now while I'm still thinking about the smps regulator as it looks like I could squeeze in onto the main pcb itself although I will still have an extra breakaway section.
    Looking nice...
    May be best to put the SMPS on the extra section, as it likely needs decent multiple decoupling caps too...
    I'm still worried about the SOT89 package being rather small for a linear regulator.
    TO-252 can be persuaded to fit, but only if the PCB increases in X by ~ 7mm - then all 3 fit across the full PCB Y nicely.

    SOT223 looks like you can just fit in a V-H-V stackup, with minor tweaks to pad stacks to nudge them into the tram-lines, and then I'd place many via-thrus from TAB areas, to copper areas on the rear.
    A SOT223 variant can be made with a narrower centre pin, to take either SOT223 or SOT89.

    If you export the interim variant above as a PCAD file, I can see if I can pack in the SOT223's ? (my older copy lacks the oval 50 mil pins)
  • Looking good. Are you going to try to fit some hash marks or fiducials on diagonal corners?
  • @jmg - the SOT223 reg is a bit of a beast and and even though its thermal resistance is under half of the SOT89 I am kinda hoping that we get some figures from Chip soon regarding what the simulations show. Besides, I have the SC202 which is smaller than the SOT89 and runs cool at 500ma. I might even allow for one in the 3V3 supply too. DN8 is also available but won't be hand solderable (put pastable) just like the SC202. The standard layout is easy enough to assemble with medium flat tip as these tips aid heat transfer and solder reflow. I may get do up a board with a dummy TQFP100 just for the exercise.
    I will look at doing a "power strip" with SOT223s perhaps.

    I've updated the exports and I will also generate some gerbers but just for reference there are layer images in a folder.
  • Cluso99Cluso99 Posts: 18,066
    Very nice layout as usual Peter !

    A few thoughts...

    I see you have separated the 3V3 I/O power rails to two supplies. Is there any way to link them for an option to use a single regulator?

    What is the main xtal? Perhaps the smaller 3225 powered osc could have a link so that you could also use a passive xtal too (ie power & gnd links). Then perhaps there would be no need for the larger xtal. Just a thought.

    BTW for the P1 I now use the CP2102 6pin for ~$1.30 on eBay and put the transistor reset on my P1 boards. This uses a 5-pin SIL header as DTR, RXD, TXD, +5V, GND (and 3V3 unused). This significantly reduces the cost entry point.
  • Peter JakackiPeter Jakacki Posts: 10,193
    edited 2018-06-11 02:25
    Cluso99 wrote: »
    Very nice layout as usual Peter !

    A few thoughts...

    I see you have separated the 3V3 I/O power rails to two supplies. Is there any way to link them for an option to use a single regulator?

    What is the main xtal? Perhaps the smaller 3225 powered osc could have a link so that you could also use a passive xtal too (ie power & gnd links). Then perhaps there would be no need for the larger xtal. Just a thought.

    BTW for the P1 I now use the CP2102 6pin for ~$1.30 on eBay and put the transistor reset on my P1 boards. This uses a 5-pin SIL header as DTR, RXD, TXD, +5V, GND (and 3V3 unused). This significantly reduces the cost entry point.

    Actually, that's a thought, to allow for those cheap USB serial modules and put a reset on-board. I could add the 5-pin to the left hand strip. BTW - when the module says TXD do they mean that it is the TXD of the module (as it should) or are they labeled for Fritzers so that it means "connect to the TXD of the other board" ? :) I suppose I don't have to allow for the actual 6-pin connection anyway since they supply them with individual jumpers.

    I use a lot of these cheap 2 pad 5x3mm smd crystals (and I have hundreds of 10MHz ones) but checking Mouser I see that there are a lot of 4 pad devices available so I will dual footprint those or just switch to a smaller footprint perhaps.

    One of the reasons for a separate regulator is to cut down on power dissipation on any one package but also to isolate the CPU and boot itself from any I/O that may be overloading the main I/O regulator.

    Please if anyone has any thoughts and you have particular parts in mind, could you please include the link as it saves having to search and guess.




  • jmgjmg Posts: 15,140
    Cluso99 wrote: »
    BTW for the P1 I now use the CP2102 6pin for ~$1.30 on eBay and put the transistor reset on my P1 boards. This uses a 5-pin SIL header as DTR, RXD, TXD, +5V, GND (and 3V3 unused). This significantly reduces the cost entry point.

    Those are ok, but they are unclear if they use the older OTP CP2102, or the new Flash CP2102N ?
    To pgm the older OTP parts, needed a extra CAP.

    The flash model allows re-pgm many times of details like Handshake and USB current options. I'd guess they eventually will filter thru to all the eBay units (once all the CP2102 are used up !)

    The PL2303HXD cables mention thesebaud rates
    12000000
    6000000 460800 134400 19200 1800
    3000000 403200 128000 14400 1200
    2457600 268800 115200 9600 600
    1228800 256000 57600 7200 300
    921600 230400 56000 4800 150
    806400 201600 38400 3600 110
    614400 161280 28800 2400 75
    
    but I've not tested one yet, to see what is sustainable.

  • jmgjmg Posts: 15,140
    I use a lot of these cheap 2 pad 5x3mm smd crystals (and I have hundreds of 10MHz ones) but checking Mouser I see that there are a lot of 4 pad devices available so I will dual footprint those or just switch to a smaller footprint perhaps.
    If doing a 4-pad generic one, I'd suggest making a variant with larger pads to support multiple sized Xtals.
    eg 3225 / 2520 / 2016
  • Peter JakackiPeter Jakacki Posts: 10,193
    edited 2018-06-11 04:06
    jmg wrote: »
    I use a lot of these cheap 2 pad 5x3mm smd crystals (and I have hundreds of 10MHz ones) but checking Mouser I see that there are a lot of 4 pad devices available so I will dual footprint those or just switch to a smaller footprint perhaps.
    If doing a 4-pad generic one, I'd suggest making a variant with larger pads to support multiple sized Xtals.
    eg 3225 / 2520 / 2016
    Any links? Footprints would be even better.


    Current layout (and schematic) with SC202A SMPS option integrated into main design.
    P2D2-CV1.png
    P2D2-SCH.png

  • jmgjmg Posts: 15,140
    Any links? Footprints would be even better.

    Attached is what I meant - this from KiCad's Crystals library, a 3225 with large pads, targets hand soldering, and the narrow spacing allows smaller xtals to fit on same pads.
    480 x 480 - 23K
  • jmgjmg Posts: 15,140
    edited 2018-06-11 05:21
    Current layout (and schematic) with SC202A SMPS option integrated into main design.

    Nice. That SC202A has a couple of keepout nodes, but I guess you could rely on the solder mask to keep clear ?
    I noticed the import of that SC202A complex footprint did not quite make it into kiCad 100%, looks like the fill areas associated with the pads got lost ? Still, most does import fine... better than I expected really.

    I did manage to squeeze in SOT223, but it is tight, and has no room for the SMPS part inside the PCB area.
    Image below shows SC202A import and SOT223 placements ...
    420 x 745 - 89K
  • There's also a simple trick to reducing heat on the 1.8V LDO, and that is to add a series diode from 3.3V. This means that instead of dropping 1.5V, the LDO only needs to drop around 0.7V. So if 1.8V was drawing 100ma that means a drop from 150mW down to 70mw. The same goes for the 5 to 3.3V supply too but it is easy to add an external diode for that. In fact the little jumper resistor I have from the serial connection allows for this.

    I believe the same die is used for SOT89 or SOT223 so it comes back down to reducing and dissipating the heat as the current handling capacity would be the same if kept within a safe temperature operating range. Having SOT89 allows larger 0805 cap footprints to be used for where it may be necessary. Although it is possible to source good 0603 caps for this, it is easier to get hold of 0805 types, including A case tantalums.

    I'm relying on the solder mask plus a silkscreen to make sure that the SC202A's exposed pads are not a problem.

    @jmg - so are you saying that if I make the crystal oscillator pads a little larger on the outside AND route pin 1 to XO on the P2 AND allow for 2 ground pads, that I can use 3.2 x 2.5s as well as 5x3. Making the pads larger on the outside will make it easier to "hand solder" but I'd still like to keep this 2 pad crystal footprint. I will still look at the options though.


  • Cluso99Cluso99 Posts: 18,066
    edited 2018-06-11 05:53
    Actually, that's a thought, to allow for those cheap USB serial modules and put a reset on-board. I could add the 5-pin to the left hand strip. BTW - when the module says TXD do they mean that it is the TXD of the module (as it should) or are they labeled for Fritzers so that it means "connect to the TXD of the other board" ? :) I suppose I don't have to allow for the actual 6-pin connection anyway since they supply them with individual jumpers.
    Yes, its really the TXD (output). 5 pins are fine - the 6th which is 3V3 is inadequate for anything so I often cut them off anyway.

    I use a lot of these cheap 2 pad 5x3mm smd crystals (and I have hundreds of 10MHz ones) but checking Mouser I see that there are a lot of 4 pad devices available so I will dual footprint those or just switch to a smaller footprint perhaps.
    I have been using the 4 pad 3225 xtals (12MHz). The powered oscillators come in the same footprint. They are small, not that expensive, and readily available. So the 3225 would be my choice. I wouldn't worry about other footprints, though it wuld be nice to have powered oscillators and unpowered xtal options.

    One of the reasons for a separate regulator is to cut down on power dissipation on any one package but also to isolate the CPU and boot itself from any I/O that may be overloading the main I/O regulator.
    Yes, realised that. However for cases where separate supplies isn't required, being able to join the two sections and only use one regulator would be a nice option.
    Currently I am powering the BeMicroCVa9 from the USB, and that is emulating the P2 so I cannot reasonably expect the P2 to be more power hungry for the code that we have been running.
  • Cluso99Cluso99 Posts: 18,066
    edited 2018-06-11 06:21
  • @"WBA Consulting" - a final PCB has to be functional first. If we are spacing components out just for component IDs then we are compromising the pcb layout. It is easy enough to work from an oversized printed overlay but PnP machines don't require anything printed on the pcb other than registrations. I've never seen a PnP machine that required component IDs and I don't think I've ever seen repair manuals printed these days for the vast majority of pcbs either. If I need to repair a board I use my oversized overlay print. How do you think you'd ever fit the text "10k 0805" in for every component on packed boards, let alone be readable? Besides I do use 0603s a lot. The old TH boards were never any problem though as there was always plenty of room for component ids.

    I guess I don't understand your opinions of reference designators. I never said that "10k 0805" should be on your board, just that you could use "103" as a value designator for a 10k part marked 103 to make it easy for assembly. Reference designator standards are very similar in CAD applications nowadays and having them in the silkscreen never compromises circuit design. I mean no disrespect, but my knowledge is based upon the 1700+ board designs from ~40 different customers I have reviewed over the last 13 years. Products from a variety of markets (automotive, industrial sensors, medical, 4k video, robotics, and industrial interfaces) follow the same standards of utilizing reference designators on their PCBAs. It's very easy to fit U1, R12, L9, etc. near even the smallest part without needing to adjusting the physical characteristics of the design. Some designers will not include silk for 0201 bypass caps under BGAs unless there is room to the side of the BGA to place the designators in pattern as the way the parts are.
    Anyhow, was just trying to offer some feedback, knowledge, and suggestions, but since this is the 3rd time I have annoyed one of the "P2 folks" with my comments, I think I will just stay clear of anymore P2 talk.

  • Cluso99Cluso99 Posts: 18,066
    edited 2018-06-11 06:38
    I always like to see a component designator (eg R10, C22) but using 0603 and 0402 parts have often meant that the designator is not as close to the part as I would like. Marking polarity is desired too. But I also understand that the designators are becoming harder to find pcb space for them. I guess it's just like the component values are no longer on the smallest parts as they just cannot be legibly printed anymore.
  • jmgjmg Posts: 15,140
    @jmg - so are you saying that if I make the crystal oscillator pads a little larger on the outside AND route pin 1 to XO on the P2 AND allow for 2 ground pads, that I can use 3.2 x 2.5s as well as 5x3. Making the pads larger on the outside will make it easier to "hand solder" but I'd still like to keep this 2 pad crystal footprint. I will still look at the options though.

    Yes, Xtals show element connected pins 1 & 3 and most do not define 2 & 4 - most look to be NC
    That said, I have seen a compact murata part in (2.00mm x 1.60mm), that has unused pins joined.

    ie you probably do not need to GND pin 4, just be able to isolate from 3v3

    Crystal price tends to track package, but that's not true so much for Oscillator modules, where the high volumes push to smaller packages, so they can be cheaper.
    eg cheapest TCXO at Digikey is this new one (1.60mm x 1.20mm)

    It's easy enough to support that, mainly it means no via under the footprint.
  • jmgjmg Posts: 15,140
    Cluso99 wrote: »
    I always like to see a component designator (eg R10, C22) but using 0603 and 0402 parts have often meant that the designator is not as close to the part as I would like. Marking polarity is desired too. But I also understand that the designators are becoming harder to find pcb space for them. I guess it's just like the component values are no longer on the smallest parts as they just cannot be legibly printed anymore.

    Sometimes that means a separate layer on printout / PDF and the PCB has less information on it.
    Searchable PDFs helps a lot here, as you can find the BOM R23 on the PCB PDF.
  • rjo__rjo__ Posts: 2,114
    Peter,

    Nice.

    My memory is failing me at the moment... but at some point there was a conversation about a flash chip that had unique serial numbers... it could have been another part... but the point was that by using the part each board would have a unique id. Does this fit anywhere in your design?

    I would plan to plug boards together with a minimum of wires and cables hanging out. I can imagine doing this with your layout and an adapter that I haven't found yet, but the power rails are going to have to be up to it:) The first step will involve 16 P2's.

    Thanks,

    Rich
  • jmgjmg Posts: 15,140
    edited 2018-06-11 07:52
    ... All the files are in my public Tachyon folder but here is a link to the P2 folder, just look in PCB and P2PCB etc although I haven't done a final export yet.

    The files import into KiCad OK, (minor issues) as you can see from my screenshots.
    I did find one database issue when generating a ECO / version control netlist however.

    Here is the netlist snippet
        (net (code 6) (name "P61R")
          (node (ref "RN1") (pin "1"))
          (node (ref "PL2") (pin "6"))
    )
        (net (code 7) (name "VUSB")
          (node (ref "R10") (pin "2"))
          (node (ref "PL2") (pin "8"))
    )
        (net (code 8) (name "P2")
          (node (ref "") (pin ""))
          (node (ref "U1") (pin "6"))
          (node (ref "CN1") (pin "7"))
    )
    

    That NET file re-imports fine, until it hits the P2 NET.
    That has U1 and CN2 connections, but the 50mil castellation footprint I think you have put down as free pads.
    Those have No RefDes and no pin numbers !!
    That means any netlist cannot pivot off the part-id, and so it must fail. (Surprised it seems to import and connect ok, via the translator)

    Can you change those to a proper 50 mil connector footprint ? - 2 strips is fine. That's also safer, as it means a nudge of a single pin is not possible.


  • jmgjmg Posts: 15,140
    rjo__ wrote: »
    My memory is failing me at the moment... but at some point there was a conversation about a flash chip that had unique serial numbers... it could have been another part... but the point was that by using the part each board would have a unique id. Does this fit anywhere in your design?

    The standard SO8 package parts can have a unique ID, so they should drop in ok.

  • rjo__rjo__ Posts: 2,114
    Thanks

    Power... I only want to use 1 supply:)
  • Peter JakackiPeter Jakacki Posts: 10,193
    edited 2018-06-11 07:54
    Even the W25Q80 that I've been using supports 64-bit unique ID. TAQOZ reports this when you type .SF to Print the Serial Flash information.
    P2D2 supports powering from USB serial cable, so 5V is fine but you can still power it directly as the 3.3V pins are available on the connector and the 1.8V LDO is powered from one of those 3.3V lines. If you use the SMPS though you would have to jumper the 5V lines across on the connector.

    @jmg - the thought had already crossed my mind in regards to creating a connector from the free pads. These were originally vias which I converted to free pads just so I could start to polish the chrome so to speak, and beef them up. However, I hate free pads anywhere on a board since they don't show up in schematics etc.

    I'm thinking of using a 12MHz crystal since the oscillator is designed for 10 to 20MHz crystals however the PLL is flexible enough that we could use 48MHz oscillators by the look of it. Those Digikey ones certainly are very cheap although we would have to run them from the 1.8V supply.

    BTW, it is normal for me to have all the hidden values on one mechanical layer neatly arranged and all the hidden designators on another mechanical layer so that all I do for manual assembly or repair is pdf print to fill a page with the board outline, the overlay, and the mechanical layer plus pads. These will be included in the final files once I get around to "polishing the chrome".

  • jmgjmg Posts: 15,140
    I'm thinking of using a 12MHz crystal since the oscillator is designed for 10 to 20MHz crystals however the PLL is flexible enough that we could use 48MHz oscillators by the look of it. Those Digikey ones certainly are very cheap although we would have to run them from the 1.8V supply.
    I think they can work from 3v3 too, they just expect 1.8v to make full use of the power-saving.
    The 48MHz clipped sine drive would be a question mark, as that depends on the Crystal Amp gain.
    There will be some upper MHz for reliable clipped sine drive - the 0.8v p-p is more low-pass-square wave, but the Osc's usually spec 25MHz or 30MHz max xtals.
    The 19.2MHz one I spec;d earlier would be a safer first try ..


  • rjo__ wrote: »
    Peter,

    Nice.

    My memory is failing me at the moment... but at some point there was a conversation about a flash chip that had unique serial numbers... it could have been another part... but the point was that by using the part each board would have a unique id. Does this fit anywhere in your design?

    I would plan to plug boards together with a minimum of wires and cables hanging out. I can imagine doing this with your layout and an adapter that I haven't found yet, but the power rails are going to have to be up to it:) The first step will involve 16 P2's.

    Thanks,

    Rich

    There are also winbond flash chips which add a layer of security, W74M series. The 32~128Mb are in same SOIC-8 (208 mil) footprint,
    https://www.arrow.com/en/products/search?cat=&q=w74m&r=true

  • TubularTubular Posts: 4,620
    edited 2018-06-11 09:16
    @"WBA Consulting" - a final PCB has to be ...
    Anyhow, was just trying to offer some feedback, knowledge, and suggestions, but since this is the 3rd time I have annoyed one of the "P2 folks" with my comments, I think I will just stay clear of anymore P2 talk.

    Andrew I think your comments are really useful and we're lucky to have someone of your experience here. I suspect these designs (well mine is anyway) are still evolving, and in no way ready for manufacture yet.

    The designator thing is interesting, I think this was more common in times gone past. These days with offshore manufacture I think its more common to strip a lot of information out. I know at work we put all the useful stuff on a 'drawing' layer that never gets sent off with gerbers, but gets printed out for inhouse aseembly. It also has history notes and 'to do list' for next revision, in addition to part numbers and alternative mfr part numbers


  • In my travels i found ftdi have released cheap (~$5.50ea) FT231 and FT234 adapters.
    In case others haven't seen them either
    https://au.mouser.com/search/refine.aspx?Ntk=P_MarCom&Ntt=197655408
  • jmgjmg Posts: 15,140
    .... It's very easy to fit U1, R12, L9, etc. near even the smallest part without needing to adjusting the physical characteristics of the design. Some designers will not include silk for 0201 bypass caps under BGAs unless there is room to the side of the BGA to place the designators in pattern as the way the parts are.
    Anyhow, was just trying to offer some feedback, knowledge, and suggestions, but since this is the 3rd time I have annoyed one of the "P2 folks" with my comments, I think I will just stay clear of anymore P2 talk.
    ? I think you are saying the same thing, two different ways.
    Any final PCB certainly will have a silkscreen overlay, and that will identify the main parts - P2, connectors ets.
    The debate is mainly around how small do you still try to label, and here you admit there is a smallest-size where designers skip the silk info.


  • Peter JakackiPeter Jakacki Posts: 10,193
    edited 2018-06-11 11:45
    @"WBA Consulting" - Sorry, don't think that by my countering you, that you are annoying anyone. This is a public forum and your input is pertinent to pcb manufacturing from a manufacturing perspective and besides I will make statements many times in the hope that somebody would actually agree or disagree, and say why, just as you have done. So thank you for your input and just know that if I felt convinced by some little statement that I would acknowledge that immediately.

    I don't understand why my pcb manufacturers question why I don't have designators but I do know that they seem to expect them because everybody else does. But designators are not up to manufacturers, they are not even up to the designer, they are up to the market and to also to manufacturing capabilities. Does the market expect designators? What are they? Are there armies of repair technicians out there demanding designators? Are the PnP machines refusing to pick n place because there are no visible designators? Can the silk screening really print that tiny stuff in such tiny places? All these questions that most people don't ask is no excuse for me to just follow along though. They also ask: Why aren't you using normal CPUs like ARM or writing in C++ or Java etc.
  • Heater.Heater. Posts: 21,230
    Peter,
    ...8mil design rules etc and DSPTH but standard components on one side only.
    Excuse my ignorance. What is this "DSPTH" of which you speak?
  • Double sided plated through hole?
Sign In or Register to comment.