HP 9825 Repair Part 7: How DRAM works (deep dive into a 4116 memory chip)
Вставка
- Опубліковано 8 лют 2025
- We got ROM working in the last episode, and now it's time to try to get the RAM working. Which will lead us to a history of early semiconductor memory, and discover how DRAM and memory refresh works.
HP 9825 repair playlist: • HP 9825 Repair
Ken Shirriff’s article on the MK 4116 DRAM memory chip:
www.righto.com/...
And another one on his reverse engineering of Intel's first chip ever, the 3101 SRAM memory:
www.righto.com/...
Our sponsor for PCBs: www.pcbway.com
Support the team on Patreon: / curiousmarc
Buy shirts on Teespring: teespring.com/...
Learn more on the companion site: www.curiousmar...
Contact info: www.youtube.co...
Music Credits: Crinoline Dreams Kevin MacLeod (incompetech.com)
What is more relaxing on a sunday morning than to watch CuriousMarc fix stuff while listening to elevator music. Love it!
oh no! a SHAMBLER
For me an relaxing Sunday evening, sitting next to a todler that doesn't want to go to sleep.
I knew in theory how DRAM works, but this is the first tangible practical demonstration that I've seen. The awesome model and explanation really helped elevate my understanding, thank you!
I remember when I was first playing around with electronics, I “discovered” that mosfets had a memory of their previous state.
It was super cool many years later when I learned that’s exactly how DRAM works!
Did you say "elevate"? I know the perfect kind of music for this! Lol.
Hehe me as well, I thought I knew but this helped a lot.
The worst part is knowing that as soon as I finish watching, I've gotta wait for the next episode.
It's a little bit like watching 'Lost' back then. You had to stop in the middle of the episode to not be a complete nerve wrack in how the latest find or situation will turn out ^^
@@nikreichel2232 I agreed to binge watch Lost with my wife when all the episodes were available from the rental store. It took us 3 weeks to do the 7 series. I must say that it has ruined EVERY series for me forever. I will no longer trust a script writer to have a brain and I have been saved from many a pointless series since them. It brought home the utter futility of series watching. The reason they are rubbish is that the script writers never know how long the run will be, they hope forever but know it is not so. They sprinkle all their best ideas in near the end of each series or episode to tempt you to watch another but in reality they are simply rationing you with circus tricks. You are just meat for their grinder and the total waste-of-time endding to Lost was the ultimate let-down in my entertainment industry experience.
Contrast this with CuriousMarc, each episode is new and refreshing, even the occasional repetition is bearable as it is witty (cf Mr Fancy-pants). The information is either cool new stuff or nostalgic old lore. Each series ends at the conclusion and a new series has a whole new world to explore. The diametric opposite of the sorry premise of Lost, a pointless ramble of 7 series to introduce the viewer to an unimaginative purgatory tale in the last 2 episodes.
Please keep on with this project.
I know it's probably really frustrating for you but we are all learning so much from your struggles - you're doing a great public service!
I don't see him giving up. It's not his nature to quit. I would imagine like the rest of us there's a time when things go so wrong you step back for a day or two, but knowing him the problem will nag at him and he'll be neck deep in troubleshooting again. I agree with everyone who's said that this is a great series.
Oh man, 40+ years later, and that DRAM timing diagram is still as relevant as ever. Except for most of the peeople nowdays tRAS, tCAS, tRP etc. are just some magic numbers you tune in BIOS to make your PC run faster.
well now you send command to the dram so the original meaning of tras/tcas is overcome nowdays.
Breakfast with Marc - it doesn’t get much better! It’s fascinating to see how younger people think of the 1970s fashions as amusing. Back then I wore pants much fancier than Mr Fancy Pants!
Each episode gets me more and more intrigued. I love the mystery, the progressive victories, and the unknown road that lies ahead...
There was a black art to getting DRAM RAS/CAS working back in the late 70s and early 80s, and every design I came across used, what I considered, a hack of analog RC delays intermingled with the TTL logic to fudge the prescriptive timing.
These were not even purpose designed monostables like 74123s, just occasional Rs & Cs placed in the middle of the logic flow to achieve an extra 5 or 10ns delay to achieve the setup and hold times, and strobes, in spec. And on the schematic, it was rarely annotated what delay was actually being aimed for.
While the Z80 had DRAM "support", all that was was a refresh counter presented on the address bus. You still had to do the dirty timing logic yourself.
Sinclair repurposed that Z80 refresh counter as part of a poor man's CRT controller in their ZX80.
You are 100% right! There is such a weird RC circuit across a NAND gate in the 9825, to delay RAS a tiny bit in order to form CAS!
Yep. The timings were way outside of the standard clocking used everywhere else. You would think tapped delay line sales would have been through the roof, but everyone figured out how to do it with sprinkled parts.
As soon as he said 4116 my heart dropped. I'm waiting on components to make a 4116 tester. This HP 9825 saga inspired me to return to an old project to bring back to life an odd Apple 2 clone with a macro keyboard that died from the power supply going bad and putting a higher voltage through the computer.
Normal dude: Just explains how DRAM refresh works and why. Marc: builds a demonstration rig, simulates leakage, all so we can have a visual manifestation of the principles of CMOS based DRAM. One of the best channels on YT.
I like the elevator music, my favorite section is learning how things work
Wow, that logic simulation tool looks so useful. Can't wait to see the next ep!
Those timing diagrams remembered me when i tried once to overclock RAM on my old PC, there is nothing better than learning the guts of eletronic devices, what they do and how they work
I tried using DRAM with an 8085 chip and support TTL some forty years ago. I failed miserably. Then, low and behold, the Z80 arrived on the scene. Problem solved, one look at the tech. spec. and it was all there. I could just attach my DRAM and the processor would handle the refresh for me. To me it was an awakening, I used that processor for quite a few years after that !
It wasn’t that trivial though since the row/column timing still had to be dealt with. All that Z80 did was generate the refresh address without the external refresh circuit having to steal/borrow the bus for that purpose. Handy and saves like 3-4 TTL chips: a counter, an octal buffer, and some discrete gates to put the address on some bus at the right time.
In 8085 you could generate refresh in software if you just wanted to see that the DRAM worked.
Very informative.
The brain power that has gone into chips, logic, hardware firmware software is unbelievable. And this is an old machine, like comparing a piston airplane to the Space Shuttle.
I don’t know chip about electronics, but I watch these and it’s so impressive.
As a computer engineer who graduated 23 years ago but has never practiced engineering since then, I really feel ashamed of myself when I hear about things in your videos and can't remember the details!
Keep the good work up Marc 🙏
It can come back with a bit of practice. Get a ham radio license. You'll thank me.
Yea, breakfast with Marc. I love the fact that the repairs are so complicated. I learn alot. I was wondering how is this possible to have the time for repair, editting a video and personal life :D Keep doing it! Thanks!
Its funny how this is like a trip back in time and for a monument Mr. fancy pants is back with us... but Mr fancy pants always lives on in my mind. Missing the 70's
Just blown away by your ability to frame this in an interesting manner and in a way where I have the tiniest understanding being an obsolete application coder with just fragments of assembler exposure.
I assume I am not the only one that can appreciate the amount of time it takes to setup for each test and making sure everything is correct. Although I have never done that much work, I can only image it takes an extraordinary amount of time to do all of that work to end up being just moments in the video. Such great work!
I learnt (or better described, "heard of") DRAM and DRAM refreshing by reading in primary school, never touch the real bare metal even today, and now I can say I know how the DRAM refreshes because of your refresh course, thanks. Many of these videos bring back fond memories and learning.
"DRAM" is when you remember some information by *continuously* repeating to yourself what you are supposed to remember
I knew about dram when it came out Marc kinda understated how draconian the refresh logic can be to understand esp. for (at the time) the noob hobbyist. Now not so much but I still don't want to engineer it.
Marc is stopping his troubleshooting periodically to refresh our memory bits by taking a video of the procedure. Sort of like what the DRAM is doing as it refreshes its row of bits.
Did you notice that Mr. Fancypants' pants have rows and columns? I wonder what is stored at those memory locations? 👍🤗
Marc, thumbs up for using Logisim evolution in your debugging. I used Logisim to simulate an 8 bit computer ala Ben Eater and sometimes use it to simulate digital circuits when I don’t fully understand them. I am just a novice at electronics so it’s neat to see someone of your skill level using similar methodologies as myself. This is the first time I’ve seen anyone on UA-cam or any other platform using Logisim as a debugging tool. Bravo. Mr. Fancy Pants will ride again!
I felt like the video was just getting started when you said it was late in the episode! Time flies when you're having fun and all that.
Every episode is a cliff-hanger. The investigation gets more and more extensive, we are exploring ancient technology and examining history lessons along with deep-dives into application. Keep up the good work Marc! I'm learning a lot, not understanding a fair bit of it, and on the edge of my seat waiting for the next episode. Thanks for sharing!
I can't help thinking that nodding away in a bleak assisted-living facility somewhere is an 85-year-old retired HP engineer who - if we could only find him - would look at those scope traces and immediately say, “Oops, looks like you blew out your ____________ … that's this one right here on the circuit diagram.” Possibly he's even wearing fancy pants!
This may be your best series yet. Right up there with the AGC revival.
Thanks for refreshing my memory on how DRAM works. Very plastic demonstration with the MOSFET. Last time I used one of these old DRAMs was when I replaced a broken 16k 4116 with a 64k 4164 in my ZX Spectrum. (Which works because they have the same 7 Bit refresh cycle, just some pin bending is required). The 4116 is somewhat fragile, they need three supply voltages, and die when these are not applied correctly.
It looks like you are heading for the same number of episodes as your AGC project 😀
Thanks for the level of details you are telling about fault finding methods; I am learning a lot here.
That little circuit and the 4116 diagram provide the most clear explanation of DRAM I have ever seen, congratulations! There are great free logic simulation tools since a long time, but setting up a simulation takes much time. Once set up, it does answer questions to understand a circuit way faster than a logic analyzer or scope, but the initial overhead is why it is typically done only in case of desperation.
Wow... you really do like that vintage computer. I admire your determination to get it working again.
Excellent demonstration of dynamic ram, Marc. I don't know why I've never thought to build such a simple circuit! I guess if you're testing mosfets and you touch the gate and it stays conducting that demonstrates the effect. The next step is to make the mental leap to using that effect as a memory.
How has no one mentioned that Logisim-evolution before, it looks awesome.
I honestly despaired that the Logisim project was orphaned without a successor almost decade ago, after I'd written some libraries for it (and some NES/SNES nonsense I was working on).
I am very pleased that is not the case.
That is an excellent demonstration of a basic bit in memory. I had an understanding already but not quite in this way. And actually this makes much more sense to me.
Incredibly comprehensive effort I must say. You'll work it out. I know it.
Fault finding leads the way to victory. Hurray.
Thanks 4 sharing. There's learning to be had from it.
That demo circuit was cool! Those 4116's were everywhere. Arcade games such as Joust had 24 of them, and ran them hard. I've seen symptoms of bit fade on them during operation, and it makes the game do some weird stuff.
And the trs-80 I think
That capacitance behavior you just demonstrated is exactly why modern day DDR memory controllers and the memories have two timings that are called tRFC (refresh cycle) and tREFI(refresh interval). Refresh Cycle represents how many cycles are spent to refresh the memory and the interval represents for how many cycles can the memory be accessed before a refresh takes turn.
While refreshing happens, the memory controller can not read or write the data. This is also why the smaller tRFC gets, the better performance you get or the bigger tREFI gets, the better performance you get. And also exactly why you need to increase the memory voltage in order to increase tREFI or reduce tRFC...
I programmed one of these in the 70s. Great machines. I hope you get this one working! I'd like to see the Logisim file you made.
Your persistence is both astonishing and amazing.
Perfect sunday. Breakfast, cup of coffee and CuriousMarc talking about DRAM refresh cycles while the elevator music is playing in the background. Doesn't get any better than that. Honestly, your explanations are superb!
Glad you are making good progress chasing the bad parts. Good thing you have modern test equipment to work with, was a lot harder back in the day when you only had a low bandwidth scope to track down bugs. Be glad your DRAM is not built from Intel M1404A dynamic shift registers like my first project. Used a two phase clock and +/- supply voltages but no ground pin. Be careful not to let the blue smoke escape from your good board.
DRAM is the kind of thing that I didn't like in digital electronics. It just seems so unnecessary to be so complicated. But alas, it really is more efficient in multiple ways. Kudos to you for diving in, and I am eager to see the next episode. I believe you stated that you attempted to check the individual gates, so it will be interesting to see what the cause of the refresh fault is. Is a chip functioning marginally? The DRAM demonstrator was very helpful - a schematic diagram would have put it over the top.
After all these years, I never really understood DRAM until I watched this video. Funny huh? Thank you so much!
Im here for the elevator music, well that and the excellent explanations.
This is become quite the adventure. Can't wait for the next episode!
This is the best content on UA-cam currently. Thank you for what you do. And love the fancy pants running joke.
Wow... J'ai jamais vu une meilleure explication et démonstration de DRAM avant. Très clair. Ça doit être grâce à la musique d'ascenseur 😂.
Cette série deviens de plus en plus passionnante à mesure que tu creuse ta tombe plus profondément 😉
J'en suis presque à souhaiter qu'il y aient d'autres alimentations HP qui grille 😈
Ok mr fancy pants ! You left us hanging.
Energizer sells a little keychain light which works in a similar manner to that circuit you are demonstrating.
I love all the elevator music parts! Don't apologize for them ;)
I couldn't help but notice around 1:48 that 010077 is very popular bus state on the odd instruction cycles for a number of instructions before the failed subroutine return. This seems a little odd to me.
Keen eye. I finally found what I think is the root cause of it a few days ago. Hopefully I’m right, but not repaired yet. Fodder for the next episode!
@@CuriousMarc so there's hope to this rabbit hole 🙏😁
That's an interesting observation Wilton.Do you think that as well as recording the processor bus accesses,the analyzer is showing the RAS counter refresh addresses also?That would be great,because effectively Marc maybe able to track the counter refreshes using the analyser.
IF SO,a big if,could it be that the counter is somehow getting stuck when it reaches that address?That address corresponds to just over the 4K memory boundary,could it be getting stuck at this section and failing to count any higher i wonder?
Rescue us Marc!!
@@CuriousMarc Marc i have been speculating, given a keenly curious observation from mr wilton,that your analyser is showing the RAS counter refresh addresses ,am i nuts to suggest that?
If so,recommendations for a good shrink would not be viewed offensively!!
Your DRAM explanation and demo should be required reading ! looking forward to next thrilling instalment and the continued adventures of ' Mr Fancypants '...cheers.
Excellent demonstration of what dynamic RAM is and basically how it works.
I'm trying so hard to keep up! I know I can't do this, but it does offer insights anyway.
OMG! Now I can't wait for the next episode! Hurry!
Dont ever stop making these videos
Oh man, I can hardly wait for the next episode!
Excellent content, Marc. Your videos are a very good way how to understand the basics of how computers work. By demonstrating how the chips and circuits work on these older devices I have a much better picture of modern computer logic. Also Ken's blog is amazing. I hope soon there will be an episode where more of your team can be present!
Drowning in a sea of gates and flip flops ! Can't wait to see whats going wrong in the DRAM refresh logic. I love this channel.
I use to have a UNIX mini computer that had 4MB's of those 4116's!!
Four huge AA to ZZ 19 inch square boards filled with 4116's and all the support logic!
the current consumption for the refresh must have been quite considerable
@@VVerVVurm it drew 12 amps on one leg and 14 on the other...
It didn't have a microprocessor, the processor was constructed from TTL chips, on two boards!
it was called a xelos X/3200 by perk and elmer. the same company that made the big mirror for the hubble telescope.
I've played with the mosfet charged floating gate thing on my bench before but never thought about it as a single bit of dynamic RAM till now.
And today’s discrete mosfets are much better than those in 6116 in terms of gate leakage - as long as you clean them well (that’s the critical aspect).
Brutal cliffhangers, but awsome episode. Merci Marc!
Education and entertainment all in one show. Thanks for doing a great job.
Awesome to see you use logisim to simulate how it should work ! It is a very nice tool
Great series. I can;t wait to see what is coming up in the next episode
I know you and your team will get this baby going again
Despite you want just to have a working machine, I enjoy a lot this series !
Oh, the memories. Using surplus 256kx8 simms from Macs to pretend they are 1Mx8 in an Atari ST. amazing that it worked, my single sided contraption. Used some 74HC chips.
Some tip for 4116 chips: they need negative voltage BEFORE positive. A lot of ZX Spectrum computers died because negative supply charge pump failed. Some early chips need negative, some later do not. You need check charge pump because of overvoltage, it may give too low small negative voltage and causing errors. It is gate polarization source only, practically under 1 mA current.
At least with the simulation software you can easily introduce a fault by tying a track low or high along the logic gate path and see if you can replicate the fault.
Thanks for great calming content . All best from Croatia!
11:51 you say toggle the RAS but from the timing diagram it is pulled low. Then you also say toggle the CAS but from the timing diagram it is then pulled low. They return high again at the end of the read-write or read-modify-write cycle.
Excellent demonstration of dram :D loved it
@12:58 Its a bit smarter than that, the Intel 3242 also has a 7 bit counter built in. That allows one to increment thru all the Rows and refresh them (for either 64 or 128 cycle refreshes)
Hello, your channel is always enlightening to me! I learned how this kind of DRAM chip works today! Thx
This is why, even on big, expensive mainframe computers, the engineers almost never bothered with board-level repair. Just swap the board and move on! But when finding a working machine is so rare, let alone a working spare board, this kind of repair is your only hope.
I imagine they still probably had to do that on prototype boards to debug glitches during development?
Nice DRAM demo.
Interesting debug. Seeing the data at 77633 go from 010101 to 010077 didn't make me think of a refresh problem, as you're seeing multiple bits become set. In my experience, seeing the same value but with a few bits moving from 1 -> 0 means a refresh issue. Looking forward to what you find.
Huh... I happen to have 8 of those 4116 chips on my bench right now (bringing a 16k TRS-80 CoCo2 up to 64k).
Still got a S100 bus memory board, 64k of DRAM.
SIXTYFOUR KILOBYTES
THAT IS INSANE
I PRAY FOR YOUR SAFETY
@@GeorgeTsiros Don’t worry; I’m wearing all the safety gear... wouldn’t want a flying byte to hit me in the eye. 😎
I have so much I should be getting on with in my own life, but why can't these episodes be at least 3 hours long. Coming down from the high of the episode makes me feel uneasy, I need more of this drug, the cold turkey endings are horrendous.
Repairs like these is why the phrase "Deep in the weeds" was coined.
the rabbit hole becomes deeper and deeper...
story time for the nerds is awsoe cant wait to see w new directions that trainer sends you neat
If I had a teacher at school as efficent and passionate as you Marc, I might've been in another industry today xD
No wonder we just throw out pc parts when they're bad! What a headache. Or in your case, a labor of love.
Yep. The fun is obviously in the learning, not so much in the repair... These machines are poorly documented, so it’s worth finding how they work - or at least that’s how I rationalize it ;-)
@@CuriousMarc Also, in this case, complete replacement boards may be hard to come by or expensive. We all know you love HP, so it's the rational thing for you any day!
This getting more intense than the Alto.
Marc, any chance you could release a longer more talky version of your troubleshooting sessions, as a "special directors cut" extended version? Some of us hard-core troubleshooters would like to follow the entire discussion! :P
I really like the Teletype series!
Keep up the good work!
How difficult would it be to build a replacement RAM board using a modern SRAM chip (or chips), to allow validation of the rest of the machine? Being SRAM you wouldn't need any of the refresh logic, which would hopefully reduce the complexity.
Great inspiration -thanks :). And now back to my pointers and references with refreshed energy - memory remains.
GuyBrush ThreepWood
EXCELENT EXPLANATION. BIG THUMBS UP.
I haven't watched this video to the end (not yet).
My current *guess* is that it's MORE than just the refresh circuit.
I say this because in this example, 23 have bits that are flipping both ON and OFF.
If it were purely a refresh issue, then I would have expected all bit errors to be in ONE direction.
From memory (if you'll excuse the bad pun), I think a 4116 would float towards all one bits if not refreshed.
However, your 'example' shows a value of 010101 changing to 010077!!!
Good observation and something I also wondered about.
me trying to restore a data general computer : I hope it won't burn when I power it
curiousmarc : let's reverse engineer each parts.
Years ago i use a diode in my Commodore 64 that change into 4 colours,so i can see its loading from floppy disc or cassette.
Green,orange,red and yellow.
The logic simulator is ultra cool.
4116 DRAMs are also pretty fragile (although it is usually the unintended absence of the -5v supply that kills them, IME)
So if the -5 goes away (failure) then the chip can be damaged?
Cliffhanger heaven for elecronic nerds.
Love it!
When you think a channel that can't get better improves itself.
What a saga and a Blast From The Past (BFTP?). I grew up on this stuff...
Anybody remembers Intel 4004?
Great video! Thank you!