I need help with visual analysis of integrated circuits

[DX-er]

I need help analyzing these two chips. Can anyone tell me if they are microcontrollers or FSM STATE MACHINES, the links to the photos are:

https://zeptobars.com/en/read/Casio-F-91W-OKI-quartz-watch

https://zeptobars.com/en/read/Casio-F-91W-quartz-watch-homage-china

[eutectique]

I would guess that it's a FSM. No need for a µC to increment several counters.

[Doctorandus_P]

Those Casio watches are extreme low power, and as the zeptobars site states, they run 7 years or longer from a single button cell. That alone makes an uC unlikely, and also considering the amount of these things made, (and being from Casio) a custom (possible FSM) solution (Maybe just a counter, with an added FSM for time setting and handling button presses etc) would be my first guess.

[amyk]

Standard cell layout. No obvious instruction ROMs nor RAMs.

[peter-h]

Many years ago I did an ASIC design (simulated in a Xilinx FPGA) for a low power RTC. It used a ripple counter, which is the lowest power implementation because you are clocking on the bit actually changing.

Even so, it was not optimal; IIRC it drew about 1uA. With a proper silicon design you can get it down to 100nA or less.

[DX-er]

1 month ago I asked Chat GPT who claimed that that chip is some 4 bit MCU with SRAM and MROM. But when I look at the picture, I can't find MROM, SRAM or CPU. It seems those big green squares on the left are instructions created by logic gates (flip flops). And on the right are analog circuits. On the Chinese chip I also can't see MROM, SRAM or CPU. I think that both chips are electronic machines (State machine - FSM). I also wanted your opinion. Thank you for the answers. If you have any more opinions, they are welcome.

[SiliconWizard]

This is very likely just pure logic, the bulk of which being counters and registers and yes, a FSM. After that you have LCD drivers, a driver for the beeper, a driver for the backlight LED, debouncing input blocks for the buttons and the oscillator. As the picture below suggests, there may also be a charge pump, probably to generate a bias voltage for the LCD.

As can be seen on the die, this is an OKI IC. Casio started to use OKI watch ICs in the 70's and while I don't know what's current in the latest CASIO watches, I suspect that they still largely use OKI ICs, at least for their simpler models.

Interestingly, you can see a (c) 2006 marking on the die itself, suggesting that this die was a respin of the original IC they used in the late 80's. This is still on a relatively old process node. Likely > 1 µm. The die is also marked "05MG137-A01", which is probably its OKI reference. Couldn't find anything much from this reference, but not too surprising as theses ICs are not "off-the-shelf" stuff.

Being powered from a CR2016, if we take the bottom of the typical capacity (about 80 mAh), 7 years of operation equate to approx. 1.2 µA of average current for the whole watch, probably including occasional use of the backlight and knowing that the LCD is operating 24/7, this wasn't bad at all in the late 80's. It's still not bad by today's standards.

[magic]

1 month ago I asked Chat GPT who claimed that that chip is some 4 bit MCU

Has anyone tried asking ChatGPT if there is anything that it doesn't know?

[peter-h]

This was my RTC + logic design

Spot the similarity

The RTC ripple counter is the logic block at the top. The main logic block is just random logic (MFM encoder and decoder for a 3.5" diskette emulating product called Smartdisk).

Almost 30 years ago...

[DX-er]

I wonder whether other watches, such as the Casio AE-1500, are built entirely from logic gates and flip-flops as a single Finite State Machine (FSM), or if they utilize an MCU with MROM and SRAM. Given features like the five alarms, a calendar valid until 2099, a timer, and dual time, I'm curious if these functions are too complex for an FSM and easier to implement with an MCU. On the other hand, if it were an MCU, it seems unlikely that it could run for 10 years on a single CR2032 battery?

[peter-h]

A perpetual calendar (like you get in 20k+ mechanical watches ) is easy to do in logic.

As are five stopwatches.

And those circuits draw power only in the logic currently being clocked, which is maybe 0.1% of all of it. But a CPU draws power all over all the time (unless the clock is stopped for say 999ms out of every 1000).

[amyk]

I wonder whether other watches, such as the Casio AE-1500, are built entirely from logic gates and flip-flops as a single Finite State Machine (FSM), or if they utilize an MCU with MROM and SRAM. Given features like the five alarms, a calendar valid until 2099, a timer, and dual time, I'm curious if these functions are too complex for an FSM and easier to implement with an MCU. On the other hand, if it were an MCU, it seems unlikely that it could run for 10 years on a single CR2032 battery?

Definitely not too complex as any watch or timer just needs a series of counters. Add a comparator for each alarm. This is very basic logic.

It's worth noting that one of the first widespread MSI integrated circuits were PMOS alarm clocks.

[DX-er]

1 month ago I asked Chat GPT who claimed that that chip is some 4 bit MCU

Has anyone tried asking ChatGPT if there is anything that it doesn't know?

ChatGPT but unfortunately other AIs as well as the specialized AI for electronics repairmen iFixit, do not understand how to recognize integrated circuits and often change their opinions. I asked them about the workings of a simple 1 euro calculator that has operations such as addition, subtraction, division, multiplication, percentage, square root and an 8-digit number memory, and the answer was that it has a 4 bit ALU, SRAM and MROM for instructions. They gave me the same answer about that Tetris game that was popular in the nineties. According to AI, Tetris also has an MCU.

[magic]

AI parrots information from its training data, so it has no realistic chance to truly know internal details of proprietary ICs that aren't talked about publicly much. A common problem with those systems is that they don't know what they don't know and just bullshit some superficially plausible response which has a random chance of being true or false.

[Smokey]

What a crazy time to have been making stuff in 1989.  I can just imagine that design conversation:

Engineer1 "Ok.  So we are going to make a digital wrist watch.  All it needs to do is keep track of the time and date.  Maybe a stop watch.   That's it.  So how do you think we should go about this?"

Engineer2 "CUSTOM SILICON!  It's the only way!"

[amyk]

1 month ago I asked Chat GPT who claimed that that chip is some 4 bit MCU

Has anyone tried asking ChatGPT if there is anything that it doesn't know?

ChatGPT but unfortunately other AIs as well as the specialized AI for electronics repairmen iFixit, do not understand how to recognize integrated circuits and often change their opinions. I asked them about the workings of a simple 1 euro calculator that has operations such as addition, subtraction, division, multiplication, percentage, square root and an 8-digit number memory, and the answer was that it has a 4 bit ALU, SRAM and MROM for instructions. They gave me the same answer about that Tetris game that was popular in the nineties. According to AI, Tetris also has an MCU.

That might be actually be right, but purely by chance.

https://old.reddit.com/r/retrocomputing/comments/ldwj8f/my_first_attempt_at_die_shot_brick_games/

http://seanriddledecap.blogspot.com/2017/07/blog-post_8.html

[DX-er]

1 month ago I asked Chat GPT who claimed that that chip is some 4 bit MCU

Has anyone tried asking ChatGPT if there is anything that it doesn't know?

ChatGPT but unfortunately other AIs as well as the specialized AI for electronics repairmen iFixit, do not understand how to recognize integrated circuits and often change their opinions. I asked them about the workings of a simple 1 euro calculator that has operations such as addition, subtraction, division, multiplication, percentage, square root and an 8-digit number memory, and the answer was that it has a 4 bit ALU, SRAM and MROM for instructions. They gave me the same answer about that Tetris game that was popular in the nineties. According to AI, Tetris also has an MCU.

That might be actually be right, but purely by chance.

https://old.reddit.com/r/retrocomputing/comments/ldwj8f/my_first_attempt_at_die_shot_brick_games/

http://seanriddledecap.blogspot.com/2017/07/blog-post_8.html

I'm surprised they use EPROM memory for mass-produced Tetris games. However, those chips are produced in millions of quantities and the cheapest way is to Diffusion MROM, "programmed" with the production of the chip itself.

The way they did it is more expensive because the EPROM memory itself is more expensive and it also takes time to program the chip with the appropriate firmware.

EPROMs store the logical state with a trapped electric charge in the memory cell. After many years or decades, it can happen that that electric charge evaporates and the firmware is corrupted and thus the chip is non-functional. On the other hand, MROM is immortal.

[DX-er]

Here is a video where an enthusiast examines a €1 calculator chip under a microscope.

​

​Visible on the chip (from top to bottom) are the LCD drivers, registers for storing numbers, and instruction logic built from logic gates which appears chaotic. At the bottom left, there are two large capacitors that are part of the oscillator circuit for the chip's clock. At the very top, I believe there are power transistors that drive the segments of the LCD screen.
​Some calculators, such as those from Texas Instruments, feature MROM with microcode, an ALU, and a small amount of SRAM.

Is this cheap 1 euro calculator an FSM (STATE MACHINE)?

​What’s unclear to me is how this cheap Chinese calculator performs calculations using only logic gates, without an ALU or MROM? Does anyone have an insight into this?

[magic]

There has to be some sort of "ALU" buried in those gates, but it is driven by an FSM which sends signals like "do multiplication" or "do addition now" when necessary.

[SiliconWizard]

Yes - using software and CPU cores everywhere sounds like a given now, but certainly wasn't so 30 years ago.

Digital design tools were still rather primitive, people weren't up to designing a custom CPU core just to build a watch or even a simple calculator. It just wouldn't have made any sense. CPU cores just weren't available like this for almost "free" using something open source. That just did not exist. And then there was the problem of power consumption. Those ASICs had to be ultra-low power.

As I mentioned earlier, Casio did *not* design watch ICs. They were buying ICs from OKI. And OKI had many customers, not just Casio. So it's not like a small team at Casio were just deciding to design their own chip for a watch, that's just not how it would happen. What they decided is they needed an ultra-low power IC to design a watch around, and there was not a ton of options other than buy that from a specialized company.

"Watch" ICs are still produced to this day. Except for the infamous "smart" watches that need recharged all the time, "classic", simple digital watches (whether they have a LCD display or mechanical display) are still made using dedicated ICs, not off-the-shelf MCUs.

I'm sure OKI still makes such ICs. In Europe, we have EM ( https://www.emmicroelectronic.com/ ), a swiss company, with a nice portfolio of such low-power ICs and the ability to design custom chips.

[DX-er]

Thank you for your responses.
I apologize if I am being persistent, but I have one more question for which I would appreciate your opinion. It concerns the Casio AE-1500. This watch features five alarms that can be configured either to sound daily or on a specific month and date (for example, June 10). It also includes a calendar programmed up to the year 2099, accounting for all leap years.
By holding down three buttons simultaneously, all display segments are activated. Afterward, pressing an additional button causes certain numbers to appear, such as the module number and an additional number (e.g., 108). On another identical model, this number is shown as 109, and I am unsure of its meaning.
In your opinion, is this implemented as an FSM (finite state machine) or as an MCU (microcontroller)? The device must calculate leap years, allow alarms to be set for a specific day of the year, and respond to multi-button combinations. Additionally, could the numbers such as 108 or 109 represent a firmware revision stored in MROM?

[amyk]

Could be a hardcoded mask revision or similar. At these small-scale niche applications, the line between a mask-programmed MCU and FSM logic gets very blurry.

[DX-er]

It appears that today's manufactured Casio F-91 watches use a different chip from the one produced in 2006.
https://zeptobars.com/en/read/Casio-F-91W-OKI-quartz-watch
This is because the chip underwent a revision between 2011 and 2013.