How dead are 4bit MCUs?

[gnuarm]

I have found a very small number of 4bit MCUs still on the market, but not as many as I expected.  Aren't they still used in coffee makers, remote controls, microwave ovens, toys, etc? 

I remember talking to a guy who designed toys, and he said they would review his designs and if there was a resistor that was not absolutely essential, they would remove it to save the 0.1 cents. 

Then there's also the power issue.  I would expect a 4 bit MCU to be more power thrifty than an 8 bit MCU, doing the very simple tasks these parts would be used for.  But... I don't know if this would be significant.  Even in a remote control, it sits idle 99.99% of the time, then the LED draws 20 mA for a brief moment. 

Anyone here using 4 bit microcontrollers?

BTW, I tried searching on "4 bit", but the search requires at least 2 characters in every word and I guess it sees "4" as a word.

[gnuarm]

I have found a very small number of 4bit MCUs still on the market, but not as many as I expected.  Aren't they still used in coffee makers, remote controls, microwave ovens, toys, etc? 

I remember talking to a guy who designed toys, and he said they would review his designs and if there was a resistor that was not absolutely essential, they would remove it to save the 0.1 cents. 

8 bit microcontrollers are usually pad limited these days.

What process node are the inexpensive MCUs being built on?  Low priced products are often made on obsolete processes because the equipment is essentially free, fully amortized.  Then there's the power issues.  MCUs are often used in very low power applications.  I believe it was around 90nm where the static power got to be very significant compared to the dynamic.  Later geometries focused on power and so got better by trading off improvements in speed. 

BTW, I tried searching on "4 bit", but the search requires at least 2 characters in every word and I guess it sees "4" as a word.

This search worked fine:

https://www.google.com/search?q=%224+bit%22+microcontroller

I was talking about the search here, in eevblog. 

But the Google search, while "working fine", doesn't pull up many makers of 4bit chips.

[jpanhalt]

Here are the first two of several Google hits:
https://www.emmicroelectronic.com/sites/default/files/products/datasheets/em6607_ds.pdf
https://www.electronics-lab.com/bit4-is-a-4bit-microcontroller-fully-programmable-with-only-three-buttons/

I searched on 4(four) bit  here, including some permutations, and got nothing.

[gnuarm]

I was talking about the search here, in eevblog. 

But the Google search, while "working fine", doesn't pull up many makers of 4bit chips.

https://www.google.com/search?q=%224+bit%22+microcontroller+site:eevblog.com

Thanks, I hadn't thought of that.  But it still doesn't pull up anything useful specifically.  I guess I'm looking for someone who uses 4-bit MCUs and knows of various suppliers of 4 bit MCUs that may not show up on Google's radar.  I've found two myself.

http://upt-ic.com/en/products.aspx?id=1

https://www.tritan.com.tw/en/Products/1/1141/4-bit-OTP-MCU

Neither site provides data sheets with any ease.

[gnuarm]

Here are the first two of several Google hits:
https://www.emmicroelectronic.com/sites/default/files/products/datasheets/em6607_ds.pdf
https://www.electronics-lab.com/bit4-is-a-4bit-microcontroller-fully-programmable-with-only-three-buttons/

I searched on 4(four) bit  here, including some permutations, and got nothing.

I know of EM Micro.  I haven't gotten a quote from them, but they don't seem to be cheaper than the 8 bit devices.  Someone has pointed to an 8 bit device that is only $0.05 at LCSC, Padauk PMS150C.  It can be run at 5V.  So even if production is 1E6 units, I'm wondering if a 4-bit MCU is useful.  Can the price be much lower? 

[james_s]

I really doubt there is any economic advantage these days for 4 bit MCUs, 8 bit parts are SO cheap and there are so many of them on the market. Economies of scale are going to make them cheap while 4 bit would be a very niche product. Heck now you often see 32 bit ARM processors in stuff like refrigerators and coffee makers, often with internet connectivity.

[coppice]

If you want to find 4 bit MCUs in current use these days you need to look to Japan. They make no economic or power consumption sense for most people, but the last time I looked they were still seem to be used in Japan. I doubt anyone is developing significant new 4 bit designs, but many decades old MCUs still see high volume use.

[SiliconWizard]

There's a matter of market. Even small gadgets theses days tend to have a lot of features, so going for a very underpowered MCU may make sense only for a very small fraction of devices.

Even a simple remote with a few buttons and nothing else, no screen, maybe just a LED, unless it's strictly IR, will require significant software if it uses some kind of "modern" radio, and for this use case, there are many ultra-low power RF SOCs available, already containing a MCU core, and the RF front-end.

Then as as been mentioned, there is a matter of die area. On any decently modern node, a simple 4-bit CPU will take a *very small* die area and the pads will take up a lot more, meaning you'd have most of the die empty. It's worse than bad cost-wise. So you'd probably have to design something with extremely few IOs (possibly less than even the 6 typical on 8-pin devices), which would make the IC, as an off-the-shelf component, only good for very, very niche applications.

So it doesn't make much sense.

While a general-purpose 4-bit MCU would not make much sense these days, using a very simple 4-bit core inside a custom IC can.

[free_electron]

Then as as been mentioned, there is a matter of die area. On any decently modern node, a simple 4-bit CPU will take a *very small* die area and the pads will take up a lot more, meaning you'd have most of the die empty. It's worse than bad cost-wise. So you'd probably have to

Not true. Most of the die size of a 4 bitter is eaten by rom and ram space. There used to be 4 bitters that were specialized in things like BCD arithmetic and lcd display driving (toshiba, sanyo , nec , hitachi). some of these are still around in things like microwaves , clocks and other stuff. They are typically applications where display functionality is numerical only (like microwaves, fridges , and stuff like that) or simple led's/lcd only. These things are made in old 0.25 or 0.13 technologis and run at very low clockspeed. either Rc oscillator or things like a 32.768KHz tuning fork (those things are not crystals). they do not use eprom but metal mask. development is done on emulators and they release straight to mask.

they are still out there.

[gnuarm]

I really doubt there is any economic advantage these days for 4 bit MCUs, 8 bit parts are SO cheap and there are so many of them on the market. Economies of scale are going to make them cheap while 4 bit would be a very niche product. Heck now you often see 32 bit ARM processors in stuff like refrigerators and coffee makers, often with internet connectivity.

Of course it will use a 32 bit chip for Internet connections.  But in a $20 coffee maker or a $10 remote control, there isn't enough margin to toss the cost difference when buying a million processors.

[coppice]

I really doubt there is any economic advantage these days for 4 bit MCUs, 8 bit parts are SO cheap and there are so many of them on the market. Economies of scale are going to make them cheap while 4 bit would be a very niche product. Heck now you often see 32 bit ARM processors in stuff like refrigerators and coffee makers, often with internet connectivity.

Of course it will use a 32 bit chip for Internet connections.  But in a $20 coffee maker or a $10 remote control, there isn't enough margin to toss the cost difference when buying a million processors.

The price of simple MCUs is dominated by the pin count. It really doesn't matter of the core is a 4 or an 8 bit. When you look at something like a remote control, though, ultra low power qualities really favoured 4 bit MCUs a few years ago. Now there are 8 bit options which can match the power consumption of those Japanese and Swiss 4 bit MCUs, based on technology originally developed for the watch industry.

[gnuarm]

There's a matter of market. Even small gadgets theses days tend to have a lot of features, so going for a very underpowered MCU may make sense only for a very small fraction of devices.

Not sure why you are making an emotional point, "underpowered" is not a technical evaluation.  Either a given part is adequate to do the job or it isn't.  They don't sell MCUs as "underpowered" or "overpowered".

Even a simple remote with a few buttons and nothing else, no screen, maybe just a LED, unless it's strictly IR, will require significant software if it uses some kind of "modern" radio, and for this use case, there are many ultra-low power RF SOCs available, already containing a MCU core, and the RF front-end.

"Significant"??  The vast majority of remotes are LEDs with some form of modulation to the data.  That's hardly a need for an 8 bit MCU.  Even a 4-bit MCU is "overpowered" for the task.

Then as as been mentioned, there is a matter of die area. On any decently modern node, a simple 4-bit CPU will take a *very small* die area and the pads will take up a lot more, meaning you'd have most of the die empty. It's worse than bad cost-wise. So you'd probably have to design something with extremely few IOs (possibly less than even the 6 typical on 8-pin devices), which would make the IC, as an off-the-shelf component, only good for very, very niche applications.

That's one reason I'm asking about this.  Some people think the minimal MCUs priced under $0.25 are made on more recent processes.  I'm thinking they use obsolete, fully depreciated equipment on processes that have very low static leakage power.  Processes that have no trouble with 3.3V and even 5V I/Os.

So it doesn't make much sense.

While a general-purpose 4-bit MCU would not make much sense these days, using a very simple 4-bit core inside a custom IC can.

I once spoke with someone who designs toys, and they scrape every fraction of a penny, including resistors that are not absolutely essential in the product.  So I'm thinking 4-bit devices are still used in such devices.

It would be inside a custom IC where the real estate is not so important.  Such chips are going to be made on more recent processes and the additional die area is not so important. 

[gnuarm]

I really doubt there is any economic advantage these days for 4 bit MCUs, 8 bit parts are SO cheap and there are so many of them on the market. Economies of scale are going to make them cheap while 4 bit would be a very niche product. Heck now you often see 32 bit ARM processors in stuff like refrigerators and coffee makers, often with internet connectivity.

Of course it will use a 32 bit chip for Internet connections.  But in a $20 coffee maker or a $10 remote control, there isn't enough margin to toss the cost difference when buying a million processors.

The price of simple MCUs is dominated by the pin count. It really doesn't matter of the core is a 4 or an 8 bit. When you look at something like a remote control, though, ultra low power qualities really favoured 4 bit MCUs a few years ago. Now there are 8 bit options which can match the power consumption of those Japanese and Swiss 4 bit MCUs, based on technology originally developed for the watch industry.

Not trying to give anyone a hard time, but I'm asking what is actually used in the products.  I'm also interested in hearing from people who have used 4 bit MCUs.  So far, most of the comments have been speculative as far as I can tell.

[wraper]

I really doubt there is any economic advantage these days for 4 bit MCUs, 8 bit parts are SO cheap and there are so many of them on the market. Economies of scale are going to make them cheap while 4 bit would be a very niche product. Heck now you often see 32 bit ARM processors in stuff like refrigerators and coffee makers, often with internet connectivity.

Of course it will use a 32 bit chip for Internet connections.  But in a $20 coffee maker or a $10 remote control, there isn't enough margin to toss the cost difference when buying a million processors.

There are Cortex M0 MCUs which are cheaper than the vast majority of the 8-bitters. Only if really digging the bottom you will find very limited 8-bitters which are cheaper than any 32-bitter, like Padauk PMS150C. Usually for the same price as normal 8 -bitter you can find 32 bit MCU with better peripherals, more FLASH and RAM, and often lower power consumption. EDIT: Though availability sucks with current chip shortage, and prices became all over the place.

[coppice]

That's one reason I'm asking about this.  Some people think the minimal MCUs priced under $0.25 are made on more recent processes.  I'm thinking they use obsolete, fully depreciated equipment on processes that have very low static leakage power.  Processes that have no trouble with 3.3V and even 5V I/Os.

Older processes are not necessarily cheaper. People aren't making fine geometry MCUs when it would be cheaper to use a coarse geometry, and most MCUs don't need to be especially fast or low power to succeed. If you want really cheap stuff, there is a narrow band of geometries where the cost is optimised at any point in time. Try looking at foundry wafer prices (if you can find any realistic ones openly available) and you'll see.

[coppice]

There are Cortex M0 MCUs which are cheaper than the vast majority of the 8-bitters. Only if really digging the bottom you will find 8-bitters which are cheaper than any 32-bitter, like Padauk PMS150C. Usually for the same price as normal 8 -bitter you can find 32 bit MCU with better peripherals, more FLASH and RAM, and often lower power consumption.

Try negotiating for a large quantity, and see if you still think that's true.

[wraper]

There are Cortex M0 MCUs which are cheaper than the vast majority of the 8-bitters. Only if really digging the bottom you will find 8-bitters which are cheaper than any 32-bitter, like Padauk PMS150C. Usually for the same price as normal 8 -bitter you can find 32 bit MCU with better peripherals, more FLASH and RAM, and often lower power consumption.

Try negotiating for a large quantity, and see if you still think that's true.

Well, for example you could get this for 9 cents @ QTY of 500 in the pre-COVID times https://www.lcsc.com/product-detail/Microcontroller-Units-MCUs-MPUs-SOCs_Nuvoton-Tech-NUC029FAE_C2640145.html

[free_electron]

Not trying to give anyone a hard time, but I'm asking what is actually used in the products.  I'm also interested in hearing from people who have used 4 bit MCUs.  So far, most of the comments have been speculative as far as I can tell.

that is going to be VERY hard to find. (both people and tools)
- the tools are very specific. it's not like you can grab a c-compiler and crank out code. it's assembly only. and the tools run on weird machinery : sun workstations, Vax/VMS, DOS and other stuff.
- to run the code you will need an emulator. for many 4 bitters there are no flash or even otp versions. Mask rom only. so you have an emulator that uses a bond-out chip. some of these bond-outs have room for a regular eprom on their back (piggyback devices). more modern emulators have the rom replaced by ram that is loadable from the host. those emulators are very expensive and you can only get them from the device makers.
- volumes are huge and development time is long. your code has to be fully debugged before you go to mask rom...

We had a 4bit core and emulated it on a xilinx FPGA board. the 'compiler' ran under dos (originally on VAX. was written in pascal for the vax). it was a macro assembler. once your code was finished the masks were made. the rom sat to one side of the chip so they would size the chip. if you needed more memory they would just increase the 'length'. it depended on volume. if you had low to medium volume they just did one rom mask. you divided the mask cost over the device , but the individual devices could have a lot of wasted rom space. every square millimeter counts. if you had large volume it was cheaper to run a full maskset because you would claim it back in silicon area. the cost of a wafer is fixed. if you can double the number of devices , you immediately halve the die cost. not that many of these devices do not get packaged ! there is no packaging cost. they simply bond the chip directly to the pcb (glob-top). so your cost is really raw silicon.

applications ?
interior light for a car. open door : light 'fades in fast'. close door : light stays on for some time , then fades out. when you unlocked using remote it would also do the fade-in thing.

blower for a car airconditioning. drive 3 phase BLDC with variable speed control.

clock/thermometer for kitchen oven. you could set temperature , start time. the entire chip was designed to run directly off 220 volts. the only external components needed were a series resistor , a small 10uF capacitor, the power relay and a tiny 600 ohm transformer to hook up the temperature probe and have it galvanically isolated ( as the chip ran off mains you needed that for the probe). the chip had a controlled rectifier on board: as soon as the line voltage start to reach like 10 volts they would open the "rectifier". basically did zero cross detection : activate the mosfet and as soon as we hit a specific point : open the mosfets. the resistor was only there for 'startup bleed'. chip directly drove a VFD. that thing was mass produced for years. the rom had several options. depending on what the customer paid for they zapped fuses during trim to disable certain features.

linear hall sensors : read the adc , do some filtering , apply trimming data that was stored during manufacturing (laser fuses) and then drive a dac to send analog signal back out.

[langwadt]

Here are the first two of several Google hits:
https://www.emmicroelectronic.com/sites/default/files/products/datasheets/em6607_ds.pdf
https://www.electronics-lab.com/bit4-is-a-4bit-microcontroller-fully-programmable-with-only-three-buttons/

I searched on 4(four) bit  here, including some permutations, and got nothing.

I know of EM Micro.  I haven't gotten a quote from them, but they don't seem to be cheaper than the 8 bit devices.  Someone has pointed to an 8 bit device that is only $0.05 at LCSC, Padauk PMS150C.  It can be run at 5V.  So even if production is 1E6 units, I'm wondering if a 4-bit MCU is useful.  Can the price be much lower?

if you could get them for free for 1E6 units you'd save 50k, how much engineering time are you going to spend cramming things into a 4bit mcu?

now if the production is 1E6 a month/week..

[gnuarm]

Here are the first two of several Google hits:
https://www.emmicroelectronic.com/sites/default/files/products/datasheets/em6607_ds.pdf
https://www.electronics-lab.com/bit4-is-a-4bit-microcontroller-fully-programmable-with-only-three-buttons/

I searched on 4(four) bit  here, including some permutations, and got nothing.

I know of EM Micro.  I haven't gotten a quote from them, but they don't seem to be cheaper than the 8 bit devices.  Someone has pointed to an 8 bit device that is only $0.05 at LCSC, Padauk PMS150C.  It can be run at 5V.  So even if production is 1E6 units, I'm wondering if a 4-bit MCU is useful.  Can the price be much lower?

if you could get them for free for 1E6 units you'd save 50k, how much engineering time are you going to spend cramming things into a 4bit mcu?

now if the production is 1E6 a month/week..

When you use words like "crammed" into, you have decided the result before looking at the data.

How much MCU/memory does it take to run a basic microwave, coffee maker, remote control?  That is just not an important consideration in the applications being discussed.

[free_electron]

many micros had 2 or 4k rom. 8049, 8051 ... that is the plague these days. code bloat.
A smart traffic light controller using a 8048 had 700 bytes of firmware !

that 4 bit micro i was talking about had a 12 bit address bit.
intels 4004 as 12 bit address and they made a complete calculator with that !

[Doctorandus_P]

Even my toothbrush uses an MSP430, while the simplicity of it's task could easily be done by a 4-bitter.

In this modern world there is no relation between what it costs to make something, and the amount of money you have to spend to obtain an object.
According to octopart, you can get the small MSP430FR2000IRLLR for 28ct in 1000+ quantities, but I'm guessing you can get them quite a lot cheaper if you skip the retailers and buy millions of them directly from the factory.

The padauk uC's is probably as low as you can get price wise. I think they went as low as 3ct, and I think even those are 8-bitters.

Another guess is that the market for 4-bitters is too small to reach really high quantities, and therefore the more universally usable 8-bitters may be cheaper.
Setting up a semiconductor factory is an expensive business, and to make a profit with 10ct parts, selling millions of them is not enough.

If a 4-bitter is only available in some old process technology and it's sleep current is too high, this may be prohibitive. Saving 5ct on a uC does not help if you need to spend an extra 10ct on a bigger battery.

Recently I saw a documentary from ASML, and they claimed that around 95% of the equipment they ever made (from the '80-ies) is still in use, and I believe that.
Once you make equipment very accurate, keep it clean (cleanroom!) and have regular maintenance, it does not wear anymore and life is virtually infinite. With a perfect oil film there is no mechanical contact, and thus no wear. It's small imperfections, oil contaminated with dust and such that cause wear. (Edit: apparently no oil in semiconductor equipment, but the general principle still stands: better fits in higher quality equipment lowers wear and increases endurance)

In the end this means a lot of the lower-tier stuff is made on quite old equipment, but I don't know how that relates to 4 versus 8 -bitters. 

[brucehoult]

What does it even mean for a CPU to be "4 bit"?

Surely if it's going to be useful then RAM addresses are going to have to be at least 8 bits, if not 12 or 16. The same with program code addresses, if they are different. Maybe I/O space can get away with 4 bits (if it is different to RAM).

Surely instructions are bigger than 4 bits too?

A 4 bit ALU? Well, there's the Z80.

4 bit accumulator?  I guess so. You can deal with booleans, decimal or hex digits.

Given all the things in it that *can't* be 4 bit, is it really that much of a savings over an 8051, PIC, even 6800 or 8080?

[free_electron]

What does it even mean for a CPU to be "4 bit"?

the ALU is 4 bit. code words are typically 8 , 10 or 12 bit. there are some very weird architectures out there. jump with offset. sometimes the instruction set is not linear.  often used instructions are 3 bit , the fourht bit acts . the fourth bit indicates these instructions need an additional bus cycle to fetch another 4 bits...
data ram is 4 bit. like i said these processors are geared towards decimal number processing. clocks, weighing scales , meters. anything where the user interface is digit based. some have A/D that measure directly in BCD.

[free_electron]

With a perfect oil film there is no mechanical contact, and thus no wear. It's small imperfections, oil contaminated with dust and such that cause wear.

Oil ? IN A FAB ?!? are you CRAZY ? it gets EVERYWHERE. And it spreads like the plague. contaminate your machines and your downtime will be horrific.

No oil in a fab. Krytox PTFE, yes. Oil ? no way ! The only oil in a fab is in the high voltage tank for the ion implanters. and that sits outside the cleanroom.