Suitable SRAM Replacement

[matthew180]

3 minutes ago, brain said:

... since I don't know that I see any reason it all has to be surface mount. ...

Size, cost, and availability or parts.  SMD parts at the scale (SSOP packages) used in this design are not that hard to work with, and anyone who can do a decent job with through-hole parts can solder a SSOP with 0.65mm pitch.  Also, the SMD parts being almost half the size of the DIP parts, the PCB could be smaller, and that will make the boards cheaper.  Using SMD also allows the option to consider automated assembly.

 

I was also reading Fred's website and he mentions there are some problems with various IDE drives and other devices like CD-ROMs, etc.  If there is a bug in the design, it seems troubleshooting that problem would be the first priority.  Why would you spend the effort to re-layout a PCB that is known to have problems?

 

From Fred's site:

Quote

While using the IDE-card I have the most success with CompactFlash cards.  Some people have found IDE harddisks which are functioning with the IDE-card ...

Some people... some drives...  I would not want to spend $200 on a PCB and parts, then have to start cycling through IDE drives to find one that seems to work reliably.

[Swim]

21 minutes ago, brain said:

I'd possibly be willing to redo the schematic and PCB and make it all through hole, but I'd need help...  If I did, I'd probably make the entire design through hole parts, since I don't know that I see any reason it all has to be surface mount.

Check with Ksarul on this as I recall he started a thru hole project for this IDE controller a year or two ago.

[brain]

Just now, matthew180 said:

Size, cost, and availability or parts.  SMD parts at the scale (SSOP packages) used in this design are not that hard to work with, and anyone who can do a decent job with through-hole parts can solder a SSOP with 0.65mm pitch.  Also, the SMD parts being almost half the size of the DIP parts, the PCB could be smaller, and that will make the boards cheaper.  Using SMD also allows the option to consider automated assembly.

 

I was also reading Fred's website and he mentions there are some problems with various IDE drives and other devices like CD-ROMs, etc.  If there is a bug in the design, it seems troubleshooting that problem would be the first priority.  Why would you spend the effort to re-layout a PCB that is known to have problems?

 

From Fred's site:

Some people... some drives...  I would not want to spend $200 on a PCB and parts, then have to start cycling through IDE drives to find one that seems to work reliably.

As a designer, I'm aware of the advantages of SMT, but I don't see the necessity in this case:

 

• It's a PEB card.  The design has tons of room, according to the gerbers on the site
• I spot checked a few of the parts.  74ALS02 is #.35 for DIP, $70 for SOIC.  Granted, the cheap part is a non stock, but active, but even the normally stocked Digikey DIP part is $.84, hardly cause for pricing concern.  The others look to follow the trend.
• All of the parts look easily available in active status at Digikey.

I offered because it looks like finding this arcane SRAM is taking up a lot of mental bandwidth, PCB design is not tough (given at least a nominally working schematic), it would seem laying out a board and having a real schematic (as opposed to this ASCII stuff on the page) would greatly help in debugging, and how would one debug without a real board to test with, and if folks would appreciate a through hole design, it seems easier to just recreate the board as TH and then debug it. (TH boards are easier to pull ICs and debug than SMT, anyway.

 

Jim

[brain]

4 minutes ago, Swim said:

Check with Ksarul on this as I recall he started a thru hole project for this IDE controller a year or two ago.

I'll defer to the OP on that.  I'm not fully invested in this, just noting some help I could offer if needed.

 

Jim

 

 

[matthew180]

17 minutes ago, brain said:

It's a PEB card.  The design has tons of room...

True, but that is the problem, a PEB card does not have to be full-size, and having it that way is contributing to the high-cost of the board itself.  Smaller parts means a smaller board, and that makes it easier and cheaper to get fabricated.

 

16 minutes ago, brain said:

I offered because it looks like finding this arcane SRAM is taking up a lot of mental bandwidth...

 

I don't know if I would call it "arcane" per se, just that 5V devices are getting harder to find in general because they are just not being manufactured any longer.  If you are going to go through all the work of a full re-layout, if you just change the 245 buffers to 3.3V devices with 5V tolerant I/O (i.e. single voltage level-shifting buffers), then you can easily change the rest of the design to 3.3V and have the access to the more modern parts that are in active status and regularly stocked at suppliers.

 

24 minutes ago, brain said:

... it seems easier to just recreate the board as TH and then debug it.

At $200 per board?  Would it not be less expensive and better to prototype the board and debug it, then do a layout based on a known working design?

[+Ksarul]

Through-hole does have the advantage that most hobbyists with general soldering skills are comfortable assembling one. It also makes replacing defective parts during troubleshooting a lot easier (assuming the builder used sockets). I did start trying to convert the board to through-hole a while back, but it is really on the deep back burner since the HRD4000B project has progressed to the point where it is viable.

 

I for one, would appreciate a through-hole respin of the board from @brain, as I really like his board designs.

[Swim]

This is a pic of the IDE controller with compact flash that has been running 24/7 for almost a decade in the Hidden Reef BBS P-Box and holds the up/down load files.

 

[matthew180]

3 minutes ago, Ksarul said:

I for one, would appreciate a through-hole respin of the board from @brain, as I really like his board designs.

 

For sure, when it comes to hobby stuff, what you want is always going to be the first design goal.  I'm not trying to be argumentative, just trying to offer some practical suggestions.  People are welcome to take them or leave them.

[Swim]

There are a few design problems with this controller card like many designs problems don't exist until the design makes it out into actual use. The 10K resistor location next to the heat sink at the bottom right corner needs to be placed away from the back edge of the card as it often gets knocked off when the card is inserted into the P-Box. I can't tell you how many of those I replaced for TIers that often swap cards in and out.

[brain]

1 hour ago, matthew180 said:

True, but that is the problem, a PEB card does not have to be full-size, and having it that way is contributing to the high-cost of the board itself.  Smaller parts means a smaller board, and that makes it easier and cheaper to get fabricated.

I'm happy to be wrong, but I don't see a way to cut that much cost out of the PCB.  You could arguably cut out the middle and top of the PEB card, but you need both sides to hit the guides, and you need to leave enough there (probably 1-2") to keep the strength and not allow the corners to be weak spots, you would still have plenty of room to use through hole. With a U-shaped PCB, you could panelize and use the interior as another PCB's "sides", but I'm not sure it'll be a huge savings in cost.  As for assembly, it doesn't really matter to me.  PCBA is $0.016 per pad, at the place I use, regardless of SMT or TH.  No difference in lead time either.

 

If I was going to spin the board, I'd put the CF connector on the outboard PCB component (not sure what you call that part), and put a 2mm 44 pin laptop IDE connector in the middle to plug in a laptop IDE drive, with an IDE 40 pin connector next to it, for a cable for the 40 pin IDE drives.  That should cover all the bases.

 

1 hour ago, matthew180 said:

 

I don't know if I would call it "arcane" per se, just that 5V devices are getting harder to find in general because they are just not being manufactured any longer. 

 

Maybe arcane is the not the best word, but the fact that we're on page 2 of a conversation about it suggests too much time has been spent on it.  As I noted last night, the existing design will work fine with the alternate 5V SRAM SMT part specified in the parts list, as it's 5V.  It's still hard to find, but if this just a small run, I am sure utsource or online components has it at a reasonable price. UTSource does, at ~$1-1.20: https://www.utsource.net/sch/TC554001A

1 hour ago, matthew180 said:

If you are going to go through all the work of a full re-layout, if you just change the 245 buffers to 3.3V devices with 5V tolerant I/O (i.e. single voltage level-shifting buffers), then you can easily change the rest of the design to 3.3V and have the access to the more modern parts that are in active status and regularly stocked at suppliers.

If I was going to go surface mount, I'd just scrap all the TTL, put in a Xilinx 9572XL or 144XL and the RAM (could do 4MB at that point, but not sure if this is a SAMS-compatible SRAM card already or not, haven't even looked at the schematic), RTC, and IDE/CF port. But, it seems like this is an ideal kit project, and TH looks very doable.  Alliance makes a nice 5V DIP 512kB SRAM: (https://www.digikey.com/product-detail/en/alliance-memory-inc/AS6C4008-55PCN/1450-1027-ND/4234586) that fits this use case perfectly.

 

 

1 hour ago, matthew180 said:

 

At $200 per board?  Would it not be less expensive and better to prototype the board and debug it, then do a layout based on a known working design?

Anymore, I just lay out the board in EAGLE, buy a 5 pack of prototype boards, solder up and debug the design, and then re-spin.  $20 or $200, doesn't matter. 

 

I believe we should agreeably agree to disagree.  I am insanely comfortable with SMT, using it in many of my products I offer for sale, but I also enjoy TH work.  I can't argue the points that SMT is smaller and it's not beyond the hobbyist to solder, but I don't see the need in this case, and the SMT parts will make lots of folks shy away from doing it as a kit, even though you rightly point out that SOIC is easy for the novice solderer.  I see the board as being huge no matter what one does with the design, so the value of smaller size of parts concerning board size reduction will be lost.  Cost is about equal.  You probably save a buck or two in parts by going SMT on the TTL, and about $2.00 using SMT RAM instead of SMT.

 

My main issue right now is that I don't have an EAGLE lbr with the PEB card dimensions and the edge connector already.  Otherwise, I probably would have skipped responding to the forum posts and just laid out the recreation, as it looks to take part of an afternoon, tops.

 

Jim

[matthew180]

4 hours ago, brain said:

I'm happy to be wrong, but I don't see a way to cut that much cost out of the PCB.

I'm not trying to make wrong or right, just having a discussion.  I suppose "right" is the way whoever does the work decides to do it.

 

4 hours ago, brain said:

... but the fact that we're on page 2 of a conversation about it suggests too much time has been spent on it.

I don't know, I've spent many many hours trying to find parts and make design decisions.  Sometimes I end up where I started, and it seemed like a waste of time, but sometimes it resolves problems or leads me to a better solution.  Since I'm not doing the design or making the boards, I'm just offering some input based on my limited experience in retro-hardware.

 

4 hours ago, brain said:

If I was going to go surface mount, I'd just scrap all the TTL, put in a Xilinx 9572XL or 144XL and the RAM

Agreed.  I did not bring that option up because I am not willing to offer the time to write the HDL right now, so it seemed like an option that was not really feasible for those working on this bulk-build.  Maybe it is though?

 

My thoughts also went down the rabbit hole of: PATA is dead, the only available devices are used and aging, so why not do a SATA board?  Oh, look, !@#$ SATA, those pricks made it all NDA and closed information licensed like HDMI...  Hmm, well, SATA is really dead now too, anyway, so why not check out M.2?  Oh, M.2 is just PCIe, and that is open, and look NVMe specs are open and available and is just a way to talk to the M.2 cards via a PCIe interface.  So why not just make a card that has an M.2 interface?  Oh, well, why not skip all that complexity and just make a card that interfaces to NAND Flash ICs directly?  The hardware would be minimal, and just write a driver to make it look/act like a disk.  Let's see, yup NAND Flash is available on DigiKey, about $10 for a *single* 8GiB chip.  Oh, well, forget that, why not just make a network card and mount the storage from a server and skip all the hassle of trying to keep up with mass storage? ...

 

4 hours ago, brain said:

Anymore, I just lay out the board in EAGLE, buy a 5 pack of prototype boards, solder up and debug the design, and then re-spin.  $20 or $200, doesn't matter.

Sure, for my own projects I tend to do that as well (KiCAD and OSHpark for me).  But a lot of that is because the parts I'm using are too small to prototype any other way.  For this IDE project I have no idea what the skills are of those trying to put the build together, and redesigning or troubleshooting the board is probably not what they were expecting.

 

5 hours ago, brain said:

I believe we should agreeably agree to disagree.

Ok.  I was not really trying to disagree though, so I apologize if it came across that way.  There are many options, and no one is any more right than any other.

 

To the original question, the main point I was trying to stress is that whatever part is chosen for the SRAM and clock, they need to be 5V parts (VCC and I/O capability) if the existing PCB design is going to be used.

[brain]

2 minutes ago, matthew180 said:

I'm not trying to make wrong or right, just having a discussion.  I suppose "right" is the way whoever does the work decides to do it.

Discussion are always good.  It just seemed like it was turning into a SMT versus TH debate. And, I'm probably not the person to evangelize the TH option, since I do most of my current designs as SMT, for which I get lambasted by the retro "purists" who claim that retro HW should only use parts from the era. Yes, those people exist.

 

2 minutes ago, matthew180 said:

I don't know, I've spent many many hours trying to find parts and make design decisions.  Sometimes I end up where I started, and it seemed like a waste of time, but sometimes it resolves problems or leads me to a better solution.  Since I'm not doing the design or making the boards, I'm just offering some input based on my limited experience in retro-hardware.

I'm perusing the 26 page F18A MK2 thread at this late hour, so I see...  It was on my list of pages to scan a few days back, but I just got to it tonight.

 

Still, much like some of the middle pages of the F18A MK2 thread), there's been lots of posts on this thread that didn't move the ball forward (mine, for example, agreeing that the replacement parts fit the original footprint, but it's a mot point, since the original part is wrong for the design.  The folks trying to justify the 3V3 coming out of the RTC as being why it's ok, when that's just another wrong path, as the IO lines will be slammed with 5V signals the SRAM can't handle. And, then this whole portion I'm probably flogging the dead horse on about a recreation and what design ethos to pick for such a task, when the OP really just wanted to know which part would work - answer: none of the ones they listed ?) The easy solution (for the OP, anyway) is to source the TC554001A part and have someone update the BOM on that page to remove the 3v3 part listed (as it's wrong, to your original point).  Hopefully we'll hear from the OP on the matter.

2 minutes ago, matthew180 said:

Agreed.  I did not bring that option up because I am not willing to offer the time to write the HDL right now, so it seemed like an option that was not really feasible for those working on this bulk-build.  Maybe it is though?

It probably depends on what the masses really want.  I stopped by because I had expertise to check datasheets, and the OP asked.  An IDE interface is a well traveled path for myself, but I don't know that I need one for my PEB right now.  On the other hand, I've been struggling to find a nice way to do something of value with the TI 99/4A I own, since I use such projects to learn the platform.  I didn't actually know this was for a bulk buy/build honestly, I just came for the datasheet ask ?

2 minutes ago, matthew180 said:

My thoughts also went down the rabbit hole of: PATA is dead, the only available devices are used and aging, so why not do a SATA board?  Oh, look, !@#$ SATA, those pricks made it all NDA and closed information licensed like HDMI...  Hmm, well, SATA is really dead now too, anyway, so why not check out M.2?  Oh, M.2 is just PCIe, and that is open, and look NVMe specs are open and available and is just a way to talk to the M.2 cards via a PCIe interface.  So why not just make a card that has an M.2 interface?  Oh, well, why not skip all that complexity and just make a card that interfaces to NAND Flash ICs directly?  The hardware would be minimal, and just write a driver to make it look/act like a disk.  Let's see, yup NAND Flash is available on DigiKey, about $10 for a *single* 8GiB chip.  Oh, well, forget that, why not just make a network card and mount the storage from a server and skip all the hassle of trying to keep up with mass storage? ...

A fair progression.  On the other hand, IDE is easy, it's still available (it seems like you can buy new drives, not sure if just NOS, or what), and for someone looking for a PEB design #1, it could be interesting.  I do agree that a design that implements a FDD and HDD controller but talks Wifi or Ethernet out the back end is a better option, but crawl, walk, run, as they say.

2 minutes ago, matthew180 said:

 

Sure, for my own projects I tend to do that as well (KiCAD and OSHpark for me).  But a lot of that is because the parts I'm using are too small to prototype any other way.  For this IDE project I have no idea what the skills are of those trying to put the build together, and redesigning or troubleshooting the board is probably not what they were expecting.

Well, my counterpoint to your note was that if I was recreating the design, I'd just design up the schematic + PCB, have the first spin, solder it up, test it out, make the bodge fixes, update the schematic, and respin a final board for those who wanted the design (I put all my stuff on github.com, so folks could pull from there).  I wasn't suggesting recreating and then just throwing the untested design online for everyone to debug.  When I do TH designs, it's just easier anymore to do up the board in EAGLE (I'd love to slide over to KiCAD, but I'm so fast on EAGLE, and learning a new schematic capture/PCB CAD tool at this point seems like a waste of time I could spend designing another idea), spin it, and test that way.  breadboards always seem to fail me at the worst moment, and I rarely find the time to hand solder up a perfboard of an entire project (I can do little adapters or real simple projects, but anything reasonably complex I forget a trace and waste time tracking it down, etc.)

2 minutes ago, matthew180 said:

 

Ok.  I was not really trying to disagree though, so I apologize if it came across that way.  There are many options, and no one is any more right than any other.

 

To the original question, the main point I was trying to stress is that whatever part is chosen for the SRAM and clock, they need to be 5V parts (VCC and I/O capability) if the existing PCB design is going to be used.

Yes, and it's a fine point.  The part referenced in the original post is, quite frankly, unworkable.  And, thus, so are the possible replacements.

 

Jim

 

 

[Shift838]

I did hear back from Thierry and he said the part number on his site is wrong and it should be be the 5v version.

 

The correct one is 'M68Z512-70NC1'

 

 

[GDMike]

Nice you were able to talk with him..that's awesome all in itself..

[Shift838]

On 3/28/2020 at 5:07 PM, matthew180 said:

Ah, yeah, right.  Too many options, too easy to make a mistake.  Even with the internal SRAM, the internal battery is going to be a problem down the road.

One of the reasons I was looking at the RTC-65271 with the external SRAM so that the batteries can be replaced easily.

[F.G. Kaal]

On 3/28/2020 at 9:59 PM, Swim said:

Hi Fred, there is supply of BQ4852YMC-85 but Rochester electronics instead of no longer available Texas Instruments. 

 

https://www.digikey.com/product-detail/en/rochester-electronics-llc/BQ4852YMC-85/2156-BQ4852YMC-85-TI-ND/11538020

 

Yes I meant the BQ4852 clock chip. This is the one with the internal RAM so you don't need the Vout and the RAM chip.

I have one IDE card running with this clock chip for year now. But it is true that when the battery is empty then there is a problem and some external voltage supply for the RAM chip must be created.

[Shift838]

2 hours ago, F.G. Kaal said:

Yes I meant the BQ4852 clock chip. This is the one with the internal RAM so you don't need the Vout and the RAM chip.

I have one IDE card running with this clock chip for year now. But it is true that when the battery is empty then there is a problem and some external voltage supply for the RAM chip must be created.

since we all know the batteries are going to eventually die and these BQ4852 chips are pricey still I am thinking maybe a better solution is to use the RTC-65271 along with the 3v batteries that are replaceable in this device.

 

Thoughts?  Yes i know we need to put the SRAM on the board.  The other thought is when the battery runs out then we need to do some dremel work to get to the battery contacts maybe to mount an external 3v battery?  I have seen it done on the original clock chip in a video or web page.

[brain]

2 hours ago, Shift838 said:

since we all know the batteries are going to eventually die and these BQ4852 chips are pricey still I am thinking maybe a better solution is to use the RTC-65271 along with the 3v batteries that are replaceable in this device.

This.  So much this.

2 hours ago, Shift838 said:

 

Thoughts?  Yes i know we need to put the SRAM on the board.  The other thought is when the battery runs out then we need to do some dremel work to get to the battery contacts maybe to mount an external 3v battery?  I have seen it done on the original clock chip in a video or web page.

I've done it twice.  It's not a simplistic affair, it's prone to mistakes, and it looks terrible in the end.  (But, it does work).  Still, if you're doing this today, with new parts, use the separate coin cell if at all possible. 

 

Jim

[matthew180]

There are a lot of clock options these days, but trying to find a solution that does not involve changing the PCB and/or software might not be possible.  I have not looked in detail, but is the RTC-65271 compatible with the clock chips used in the PC since the PC-XT?  I.e. the DS12885 family?  I made a board to replace the clock+battery solution for my retro-PC computers and it worked out really well.  You can use the attached coin-cell holder, or snap-off the battery holder to place it somewhere else and use the JST header to connect it back to the board.

 

Anyway, maybe a PCB like this is the answer.  Get a clock chip, SRAM, level shifters, voltage regs, etc. all on a small PCB that can plug into the clock location on the original board.

 

But, like Jim and I talked about earlier, you can go around and around on these options forever.  You (OP) need to specify the limitations and what you are willing and able to do to make this happen.  We have no idea if PCB changes are an option, or what?  From my perspective, the design needs to be updated to accommodate the ICs available today, but I'm not willing to offer the time to do that right now, so it is just an opinion.

 

[RickyDean]

       Wow, I keep resurrecting these old threads. On the IDE card, as discussed by @Swim there is the BQ4852 that has 512k Sram on board, so if you use it you don't need the 512k sram chip. It has 36 pins two of which are not connected, one that is a programmable interrupt, and then it has a separate  Microprocessor reset, apart from A18 address line.

       On the other hand I have in my procession about 20 or so Dallas DS1251Y timekeeper SRAM 512 x 8 that is 32 pins. It does not have a programmable interrupt, and the Microprocessor reset pin is shared with A18 address. It seems like I can build a small board here and use the Dallas chip in place of the BenchMarq chip. Is the interrupt important here on the IDE card?

       The 32 pin BenchMarq BQ4842Y that can be used on the IDE card doesn't share quite the same pin footprint. I think it would be easier to fit the Dallas in the 36 pin slot. The interrupt is probably used for CRU purposes according to the schematics, so could that be done another way? What do y'all think? I'm just wanting to see if I can use what I have on hand.

 

  

Edited September 28, 2023 by RickyDean
spelling, added content

[Shift838]

36 minutes ago, RickyDean said:

       Wow, I keep resurrecting these old threads. On the IDE card, as discussed by @Swim there is the BQ4852 that has 512k Sram on board, so if you use it you don't need the 512k sram chip. It has 36 pins two of which are not connected, one that is a programmable interrupt, and then it has a separate  Microprocessor reset, apart from A18 address line.

       On the other hand I have in my procession about 20 or so Dallas DS1251Y timekeeper SRAM 512 x 8 that is 32 pins. It does not have a programmable interrupt, and the Microprocessor reset pin is shared with A18 address. It seems like I can build a small board here and use the Dallas chip in place of the BenchMarq chip. Is the interrupt important here on the IDE card?

       The 32 pin BenchMarq BQ4842Y that can be used on the IDE card doesn't share quite the same pin footprint. I think it would be easier to fit the Dallas in the 36 pin slot. The interrupt is probably used for CRU purposes according to the schematics, so could that be done another way? What do y'all think? I'm just wanting to see if I can use what I have on hand.

 

  

one of the issues that Fred discovered around the BQ4847 was that sometimes it would throw an interrupt and hang the TI system on startup.  (if I remember correctly).  His solution was to disconnect the interrupt pin (#5).  I have done that in my design by default with the BQ4802LYDW RTC, which is a direct replacement for the BQ4847 but in a SMD package.  But like I said that was on the BQ4847, @F.G. Kaal probably has better insight on the other RTC's that had built in SRAM.

 

The DS1251Y looks like it could be direct drop in replacement for the BQ4842, as far as pin count and signals, unless I'm reading the datasheet wrong.  It's still quite a bit more expensive than the circuitry required for the BQ4802 RTC.

Edited September 28, 2023 by Shift838

[RickyDean]

8 minutes ago, Shift838 said:

one of the issues that Fred discovered around the BQ4847 was that sometimes it would throw an interrupt and hang the TI system on startup.  (if I remember correctly).  His solution was to disconnect the interrupt pin (#5).  I have done that in my design by default with the BQ4802LYDW RTC, which is a direct replacement for the BQ4847 but in a SMD package.  But like I said that was on the BQ4847, @F.G. Kaal probably has better insight on the other RTC's that had built in SRAM.

 

The DS1251Y looks like it could be direct drop in replacement for the BQ4842, as far as pin count and signals, unless I'm reading the datasheet wrong.  It's still quite a bit more expensive than the circuitry required for the BQ4802 RTC.

Understood  about the expense, but, I already have them on hand. Got them at a good price a few years ago from ebay on a deal. Hate to see them wasting away. I'd need to run A18 to the reset pin and run A17 to the int pin, to fit the chip into the 4842 slot, but that could be an option too. I'll wait for @F.G. Kaal to respond, thanks Chris.

[F.G. Kaal]

13 hours ago, RickyDean said:

Is the interrupt important here on the IDE card?

The interrupt is not important on the IDE card.

 

One of the first things that the INIT routine of the IDE DSR does is to disable the interrupt of the clock chip.

One of the things I noticed in the very beginning of the IDE card is that sometimes when the DSR is not loaded yet in the RAM, and thus the RTC interrupt is not disabled, that the system hangs because the RTC had generated an interrupt after power on. That is why I disconnected the interrupt pin. Al is fine after the DSR is loaded.

 

[F.G. Kaal]

12 hours ago, Shift838 said:

But like I said that was on the BQ4847, @F.G. Kaal probably has better insight on the other RTC's that had built in SRAM.

The interrupt problem I noticed was not always after powering on with no DSR loaded, but sometimes. That sometimes is "the hell" for people with no or little hardware knowledge. Beter prevent that and leave that interrupt pin not connected.

 

[F.G. Kaal]

I still like the last RTC solution you made with the external battery @Shift838. The battery replacement is simple and you don't need a Dremel.