Why faster MCU's always get's large pin packages!

[MT]

Why does faster MCU's always get bigger pin packages (more peripherals argument exluded) and not smaller?
Why cant i buy a 1Ghz/32bit MCU in SO16?

[rstofer]

Count the number of pins used for power.  On the faster devices, I would expect to see more pins used for the various voltages.  As well, speed causes heat and the package has to get the heat out.

[wraper]

Why cant i buy a 1Ghz/32bit MCU in SO16?

Silicon die probably won't fit in the package. And what's the point in such MCU, what are you going to do with it with given pin count limitation that would require 1Ghz/32bit?
EDIT: that said without mentioning other issues like power delivery/decoupling. Also 1GHZ is not even in MCU league at all. Those SoC chips normally have external RAM and flash, often as second BGA package on top of them.

[coppice]

Do you have some high volume applications for a 1GHz MCU with very very few I/O pins? If you do, I'm sure you'll inspire some MCU maker to produce you a part in a fairly low pin count package. It probably won't be 16 pin, as they and unlikely to achieve good signal integrity with so few power pins, but they could probably do a nice small 30-40 pin BGA for you.

[brucehoult]

The 320 MHz FE310-G000 RISC-V is in a QFN48 package.  Any advance on that?

Yes, it would be nice. Sometimes you want compute more than huge I/O.

[brucehoult]

The 320 MHz FE310-G000 RISC-V is in a QFN48 package.  Any advance on that?

Yes, it would be nice. Sometimes you want compute more than huge I/O.

I found some lowish pin count chips.

64 pins 300 MHz ATSAMS70J19A-AN ATSAME70J21A-AN and friends (ARM)

64 pins 500 Mhz 8 core (!) XL208-128-TQ64-C10 XLF208-128-TQ64-C10 XU208-128-TQ64-C10 XUF208-128-TQ64-C10
48 pins 500 MHz 6 core XS1-L6A-64-TQ48-C5 (and 400 MHz 4 core XS1-L4A-64-TQ48-C4)

In fact these latter chips share the available 400 or 500 MIPS of a single core between up to 8 hardware threads, with a maximum of 100 MIPS for each thread

[NorthGuy]

You certainly can pack lot more than 16 pins into a package of the same size as SOIC 16. People who're after small size will never use SOIC.

[coppice]

You will find a number of 50-500MHz devices in fairly small packages, targeted at control applications. Many control applications only need a handful of signals, so 30 or 40 pins is quite a lot for these devices. They tend to top out at around 500MHz, though. If the device is reasonable efficient, current high volume control applications rarely need faster processing than that.

[Mechatrommer]

Why cant i buy a 1Ghz/32bit MCU in SO16?

And what's the point in such MCU, what are you going to do with it with given pin count limitation that would require 1Ghz/32bit?

on top of my head... simple hobbiest level UAV, that only control few servos, motors and sens few sensor, 3 axis gyro maybe, but inside the cpu is doing complex/matrix/floating robust adaptive control algorithm... having 64 pins cpu on board with only 16 pins used is feel like blasphemy..

[Jeroen3]

You certainly can pack lot more than 16 pins into a package of the same size as SOIC 16. People who're after small size will never use SOIC.

Exactly. You can get 80 MHz Cortex M4F in 3x3mm WLCSP49 (wafer level chip scale package).
It's an minuscule package, literally the chip flipped over with solder balls directly on an extra metal layer.

[MT]

I assumed you would see the "figuratively" metaphorically speaking , so keyword are still "small pin count+ very high speed", whether it's SO16, SSOP10 TQFP20,WLCSP10, etc, etc.

OK  so why cant i buy a SSOP10/WLCSP10/TQFP25 1Ghz 32 bitter?

[MT]

Why cant i buy a 1Ghz/32bit MCU in SO16?

And what's the point in such MCU, what are you going to do with it with given pin count limitation that would require 1Ghz/32bit?
EDIT: that said without mentioning other issues like power delivery/decoupling. Also 1GHZ is not even in MCU league at all. Those SoC chips normally have external RAM and flash, often as second BGA package on top of them.

Sometimes, actually many times you want to compute a lot on a very tiny space, audio one among many apps, serial in serial out on TDM. Other things could be distributed computing instead of one big fat complicated expensive PCB one uses many simple PCB. The argument that just because a MCU is fast it need many pins is inadequate.

Example situation: A mixed signal board, many timers, many uarts, many ADC channels etc, all runs in low power mode slow speed, CPU runs at say 27Mhz all mounted on "single side" (like swear in church) ultra low china cost PCB.

Now you need big fat powerful computing! What do you do? Slabbing on current market solutions is going to ruin the
whole solution so if there was a 1Ghz 32 bit SO8/WLCSP10 MCU (1.34euro) with a serial com's for offload that would be splendid you dont agree? Easier to say no instead?

[mdszy]

I assumed you would see the "figuratively" metaphorically speaking , so keyword are still "small pin count+ very high speed", whether it's SO16, SSOP10 TQFP20,WLCSP10, etc, etc.

OK  so why cant i buy a SSOP10/WLCSP10/TQFP25 1Ghz 32 bitter?

I think it's already been made clear that 1GHz is pretty unreasonable for a MCU.

[coppice]

I assumed you would see the "figuratively" metaphorically speaking , so keyword are still "small pin count+ very high speed", whether it's SO16, SSOP10 TQFP20,WLCSP10, etc, etc.

OK  so why cant i buy a SSOP10/WLCSP10/TQFP25 1Ghz 32 bitter?

I think it's already been made clear that 1GHz is pretty unreasonable for a MCU.

Its not unreasonable. Its just the market which is questionable. If there is a demand you could make a nice 1GHz MCU, running entirely from RAM, booted from a very cheap serial flash chip.

[NorthGuy]

Now you need big fat powerful computing! What do you do? Slabbing on current market solutions is going to ruin the
whole solution so if there was a 1Ghz 32 bit SO8/WLCSP10 MCU (1.34euro) with a serial com's for offload that would be splendid you dont agree? Easier to say no instead?

Of course, it would be wonderful if you could get nearly infinite amount of computing power in a very small package which would consume few nA when it runs at full speed and would cost $0.06. But until such things get available, you need to compromise.

Either you get a big fat processor which is big (plus lots of other stuff that it requires on PCB), consumes lots of power and costs a lot, but runs fast and gives you lots of resources to waste.

Or, you use very small low power MCU which costs little, but doesn't have any resources to waste. So, you need to apply your time and your brain to make it work.

Or anything in between.

The choice is always yours.

[MT]

If there was a choice of a 1Ghz/32 bit in a xxxxx package i guarantee that you would in less then 5 secs orders 3 samples of it! So there are no "real" choice's, only expensive ""forced-upon-you" solutions. Why are mobile phones small and dont incorporate "your big and fat" solution?

So, the "fastest+smallest" and easy to get dev software solution as of today is XS1-L4A-64-TQ48-C4.

[NorthGuy]

Why are mobile phones small and dont incorporate "your big and fat" solution?

They're not small by any means. You wouldn't have any problems finding a very fast CPU which would fit into a phone form factor. Small CPUs are much smaller than that - you can fit a dozen on a penny.

[CM800]

So.... what do you need it for? Can't really come up with many applications that are high enough volume for it to be worth while.

If it's that valuable, you could buy a wafer-level package and make your own tiny SoC for the application.

[mdszy]

If there was a choice of a 1Ghz/32 bit in a xxxxx package i guarantee that you would in less then 5 secs orders 3 samples of it! So there are no "real" choice's, only expensive ""forced-upon-you" solutions. Why are mobile phones small and dont incorporate "your big and fat" solution?

Mobile phones are using full-fledged SOCs that are optimized for size. The entire chipset is in a single package, so there don't need to be as many pins exposed to the PCB itself since lots of the connections are between dies in the same package.

[mikerj]

on top of my head... simple hobbiest level UAV, that only control few servos, motors and sens few sensor, 3 axis gyro maybe, but inside the cpu is doing complex/matrix/floating robust adaptive control algorithm... having 64 pins cpu on board with only 16 pins used is feel like blasphemy..

A Cortex M4 based micro is already powerful enough to run a flight controller for such a vehicle.

[colorado.rob]

A Cortex M4 based micro is already powerful enough to run a flight controller for such a vehicle.

It depends on how complex you flight controller will be.  But if you're using vision systems, lidar, or other fancy inputs, you need more pins anyway.  So, yeah... what @miker said.

The thing is, there is never the "perfect part".  Engineering is the art of compromise and trade-offs.  What's the cheapest thing that meets your needs?  The only reason a bigger pin package is a problem is the board space it takes up for TQFP or the number of layers and PCB cost associated with BGA.  But with a BGA, if you only need a few of the pins, routing becomes easy and PCB cost comes down.

[westfw]

Silicon die probably won't fit in the package.

Indeed.  A 1GHz cpu depends on a lot of logical structures - L1 cache, L2 cache, instruction pipelines, branch prediction units, and so on,  that are "very large" wrt the die area they occupy.  And then they eat a lot of data, so they need a lot of memory, which means more die area for either on-chip memory or memory controllers.

The limiting factor for a lot of the faster microcontrollers seems to be flash memory access speed, rather than cpu clock speed...

[fcb]

If there was a choice of a 1Ghz/32 bit in a xxxxx package i guarantee that you would in less then 5 secs orders 3 samples of it! So there are no "real" choice's, only expensive ""forced-upon-you" solutions. Why are mobile phones small and dont incorporate "your big and fat" solution?

So, the "fastest+smallest" and easy to get dev software solution as of today is XS1-L4A-64-TQ48-C4.

No point getting upset.  It's the simply the market, if there was a ready market for the size/pin count you want then i'm sure (bounded by the laws of physics etc..) that someone would be making it. Most IC's these days are designed for a lead customer (automotive/consumer/perhaps industrial sometimes).

[coppice]

Silicon die probably won't fit in the package.

Indeed.  A 1GHz cpu depends on a lot of logical structures - L1 cache, L2 cache, instruction pipelines, branch prediction units, and so on,  that are "very large" wrt the die area they occupy.  And then they eat a lot of data, so they need a lot of memory, which means more die area for either on-chip memory or memory controllers.

A 1GHz part is going to be made in a fine geometry. You can get a lot of logic on a small die at 45nm. A single core and, say, 512k of RAM isn't that big. Usually, for MCU type designs, you end up with a pad limited die. If the I/O requirements are very limited the pad ring won't be an issue, and you can make something very small.

The limiting factor for a lot of the faster microcontrollers seems to be flash memory access speed, rather than cpu clock speed...

Flash is really slow. Even with some caching and pipelining it still becomes a major speed handicap. A 1GHz device would likely boot from a small serial flash chip, and run from RAM. Even some MCUs running below 100MHz are doing this.

[KL27x]

If someone offered such a chip, even if it saved space on the pcb (which if you can operate with just the internal RAM, let's assume for sake of argument that it can be significant)... what makes you think this chip would be cheaper?

[wraper]

Sometimes, actually many times you want to compute a lot on a very tiny space

In those cases you don't use SO packages as it'a a waste of space by itself. Large package that could fit only a tiny die inside of it.

example situation: A mixed signal board, many timers, many uarts, many ADC channels etc, all runs in low power mode slow speed, CPU runs at say 27Mhz all mounted on "single side" (like swear in church) ultra low china cost PCB.

Now you need big fat powerful computing! What do you do? Slabbing on current market solutions is going to ruin the
whole solution so if there was a 1Ghz 32 bit SO8/WLCSP10 MCU (1.34euro) with a serial com's for offload that would be splendid you dont agree? Easier to say no instead?

And in such case 1 layer PCB would be inadequate even by signal integrity standpoint. No it would not be splendid. It would be a bulky piece of crap which you could put on tiny multilayer board instead. And you don't put WLCSP10 on a cheap single layer board.

[wraper]

I assumed you would see the "figuratively" metaphorically speaking , so keyword are still "small pin count+ very high speed", whether it's SO16, SSOP10 TQFP20,WLCSP10, etc, etc.

OK  so why cant i buy a SSOP10/WLCSP10/TQFP25 1Ghz 32 bitter?

WLCSP10 is too small and would be useless with pin count this low. But you can get ARM MCU with larger pin count with still tiny size like 3x3 mm (not 1 Ghz beasts of course).
TQFP25 would be large and plain stupid.

[wraper]

Why are mobile phones small and dont incorporate "your big and fat" solution?

You made me lough. They do incorporate chips with hundreds of pins.
EDIT: BTW they do have billions of transistors on the die. https://venturebeat.com/2017/01/03/qualcomms-snapdragon-835-will-debut-with-3-billion-transistors-and-a-10nm-manufacturing-process/

[Rasz]

someone already mentioned XMOS XCore at 400-1000MHz (divided by hardware threads to help with memory latency)

And what's the point in such MCU, what are you going to do with it with given pin count limitation that would require 1Ghz/32bit?

on top of my head... simple hobbiest level UAV, that only control few servos, motors and sens few sensor, 3 axis gyro maybe, but inside the cpu is doing complex/matrix/floating robust adaptive control algorithm... having 64 pins cpu on board with only 16 pins used is feel like blasphemy..

we already have those, and you dont need more than ~xx MHz for what you listed. The real computation bottleneck shows up after throwing vision into the mix, but then you need all those additional pins for video IO anyway.

The 320 MHz FE310-G000 RISC-V is in a QFN48 package.  Any advance on that?

346 MHz ESP8266 32-pin QFN ;-)  https://github.com/cnlohr/nosdk8266
and you get properly fast serial IO at 173 MHz I2S

I win?

[brucehoult]

someone already mentioned XMOS XCore at 400-1000MHz (divided by hardware threads to help with memory latency)

Yeah, that was me too. I hadn't previously heard of them until I looked for low pin count MCUs on element14 for this thread. Interesting.

The 320 MHz FE310-G000 RISC-V is in a QFN48 package.  Any advance on that?

346 MHz ESP8266 32-pin QFN ;-)  https://github.com/cnlohr/nosdk8266
and you get properly fast serial IO at 173 MHz I2S

I win?

Only until I get my order for an Arduino Cinque with one each of the above! (plus an ARM just purely so they can have USB serial&JTAG while giving evil FTDI the boot). What we don't know is how many of those CPUs will be end user programmable -- the FE310 is intended to be the application processor.

[Nerull]

If there was a choice of a 1Ghz/32 bit in a xxxxx package i guarantee that you would in less then 5 secs orders 3 samples of it! So there are no "real" choice's, only expensive ""forced-upon-you" solutions. Why are mobile phones small and dont incorporate "your big and fat" solution?

So, the "fastest+smallest" and easy to get dev software solution as of today is XS1-L4A-64-TQ48-C4.

Well, if a few hobbyists want to order free samples I'm sure the chip fabs are just chomping at the bit to spin up production. How  many chips do you need to produce to break even, 5?
 

[MT]

You made me lough.

If i made you lough i won!
http://www.thefreedictionary.com/lough

They do incorporate chips with hundreds of pins.

The point was  fewer pins not hundreds of balls!

EDIT: BTW they do have billions of transistors on the die. https://venturebeat.com/2017/01/03/qualcomms-snapdragon-835-will-debut-with-3-billion-transistors-and-a-10nm-manufacturing-process/

That just screams OEM and NDA and million volumes and 8 layer PCB, no thanks, i want 1Ghz,32Bit, 4MB SRAM TSSOP20.
Why is it so that larger the Flash/SRAM are the more pins a device get?! Surely there are very little die size difference between 32k and 64k for any same given 64TQFP package etc. Thats also like number of pins irrational and foolish marketing strategy.
Yet ST makes QFP32/TQFP48/TSSOP20 "market correct" packages utterly wrong according to some folks reasoning here.

No point getting upset.  It's the simply the market, if there was a ready market for the size/pin count you want then i'm sure (bounded by the laws of physics etc..) that someone would be making it. Most IC's these days are designed for a lead customer (automotive/consumer/perhaps industrial sometimes).

Im not upset, its some of the log conservative thinking engineers in this thread thats upset over the idea of 1Ghz 32bit 1MB SRAM TSSOP20. Speaking of the market i bet nobody understands ST market approach with their line of MCU's that make no sense anymore , very few would be able to select the "right device" for their product based on price reasons or peripherals it barely make sense in volumes.

I recently read that on a new car these days 73% of its price are electronics/software rest is metal. So what MCU and car manufacturer do is to "invent a problem" as an excuse to invent a solution to that problem, yet car manufacturers overproduce cars.

In those cases you don't use SO packages as it'a a waste of space by itself. Large package that could fit only a tiny die inside of it.

I'm talking about pin count, glad you confirms die size so make it BGA8,  1mm by 1mm.

And in such case 1 layer PCB would be inadequate even by signal integrity standpoint. No it would not be splendid. It would be
a bulky piece of crap which you could put on tiny multilayer board instead. And you don't put WLCSP10 on a cheap single layer board.

Nope it would be elegant, tiny utterly cheap and cost effective compared to your 8 layer 1million balls solution not even mention how
impossible it would be to sell it for 15 usd. But heey, "money" is an artificial figure at the end of the day.

[wraper]

Yet ST makes QFP32/TQFP48/TSSOP20 "market correct" packages utterly wrong according to some folks reasoning.

Then look again what ST offers in in those packages and reasoning makes a lot of sense.

Nope it would be elegant, tiny utterly cheap and cost effective compared to your 8 layer 1million balls solution not even mention how
impossible it would be to sell it for 15 usd

Raspberry pi 0 - $5

I'm talking about pin count, glad you confirms die size so make it BGA8,  1mm by 1mm.

If you want some 1 logic gate IC, then sure but not MCU.

Nope it would be elegant, tiny utterly cheap and cost effective

Seems you have 0 understanding about real mass production.

[MT]

Then look again what ST offers in in those packages and reasoning makes a lot of sense.

Nope it does not, just take a look at the entire production, it has all become "Michrochiped" you argue against yourself wraper!

Raspberry pi 0 - $5

A device thats old and closed for any except OEM. I also got some Nucleo boards for free.

If you want some 1 logic gate IC, then sure but not MCU.

Thats just dumb answer, certainly you cant imagine ATTINY5-TSHR SOT-23-6 is real.

Seems you have 0 understanding about real mass production.

To the contrary, but you lack imagination and future product market ,better to be stock conservative!

[wraper]

Then look again what ST offers in in those packages and reasoning makes a lot of sense.

Nope it does not, just take a look at the entire production, it has all become "Michrochiped" you argue against yourself wraper!

I don't quiet understand what do you mean by that. Where I did contradict with myself? Of course relatively slow and low current consumption cortex M0 makes sense on 2 layer PCB.

Raspberry pi 0 - $5

A device thats old and closed for any except OEM. I also got some Nucleo boards for free.

Open source has nothing to do with board production cost.

[NorthGuy]

The point was  fewer pins not hundreds of balls!

If size is important to you, then BGA might be smaller despite it has more pins. Like 121-pin 0.8mm pitch BGA is 9x9mm - it's roughly the same size as SSOP-20 (which is 8x10mm) but has 6 times more pins. If you don't use most pins on the BGA it will also be easy to fan out.

If you actually want the smaller number of pins regardless of the size, then perhaps getting a bigger package and cutting extra pins off would help.

[wraper]

Seems you have 0 understanding about real mass production.

To the contrary, but you lack imagination and future product market ,better to be stock conservative!

OK, then let's return to the through hole stuff  , to be super conservative. Industry does not agree with you for some (known) reason, hence you started this tread in the first place.

[DaJMasta]

If you go through hole, even if there's a hobby market for it, you completely cut out any potential for the package to be a commercial success.  You also limit your top chip speed because of the physical distance between contact points on the board - parasitics eating up your signal integrity margins even before you make it to the board.

I think if you want high speed and capability in a dip package, the MUCH better choice is to do something like an arduino nano module - a small board with the chip and some supporting bits with some of the pins broken out in a dip package form factor.



I think for most applications a higher pin count package doesn't really have detriments... and then it makes the same sku usable for high I/O count applications too.  While it may mean you can't hand solder it and can't use the cheapest board fabrication services... since it doesn't take a large batch before those costs are almost completely overcome, I can't see a big market for most of the things these low pin count high pin pitch fast MCUs would come to be used in.  Yeah, for me, a hobbyist, it could be nice.... but not for most applications, so can you really expect a company to cater it?

[Sal Ammoniac]

I think if you want high speed and capability in a dip package, the MUCH better choice is to do something like an arduino nano module - a small board with the chip and some supporting bits with some of the pins broken out in a dip package form factor.

This. Some people just don't understand that the hobby market is so minuscule and the numbers are so low as to make it pointless for any manufacturer to cater to this market. The only reason I can think of is perhaps for publicity (e.g. Atmel probably got some positive publicity when the Arduino people used the AVR). Other than that, the hobby market is less than peanuts to a chip maker.

Companies like Sparkfun and Adafruit can come to the rescue here by creating and selling breakout boards for chips that are otherwise impractical for most hobbyists to use.

[alm]

Not just publicity, also familiarity with your parts for possibly future EEs. If someone worked with Arduino in high school, they become an EE, and have to select a micro for a new product, Atmel (Microchip) may very well be the first place they look. That is why plenty of micro vendors tried to emulate the Arduino with the various Arduino-compatible boards and mbed. The whole slew of cheap development boards with free development environments and compilers.

However, I would not expect them to introduce a part just for that hobbyist market. They would rather stick an existing part on a board and sell that to hobbyists at a loss.

[westfw]

Quote

Surely there are very little die size difference between 32k and 64k

I think you're wrong.  Especially if you're talking about fast RAM of the sort you might want to connect to a 1GHz CPU.

I'm having trouble finding an analyzed die photo for a modern ARM chip, but I did find this one of a somewhat older ARM9 chip.  Where it shows that 4k of cache memory occupies about the same die area as the ARM core...

(transistors have gotten smaller, and I think ARM cores have gotten smaller, but the ratio of memory size to core feature size should be about the same...)

[westfw]

Other sites tout the cortex M0 core as consisting of about 12k gates.  A static memory cell is probably somewhat similar in size to "a gate", so the Cortex M0 core is about the same size as 2kbytes of RAM.

[wraper]

STM32F103VGT6 is one of the largest STMicroelectronics's Cortex-M3 microcontrollers.
1Mb of flash and 96kb of SRAM consumes most of it's enormous 5339x5188 µm die.

[CM800]

STM32F103VGT6 is one of the largest STMicroelectronics's Cortex-M3 microcontrollers.
1Mb of flash and 96kb of SRAM consumes most of it's enormous 5339x5188 µm die.

Got to love how it looks like massive fields then in the right hand bottom corner, a small town with roads, factories etc.

[Zbig]

Anyone remember this: https://www.eevblog.com/forum/projects/dawn-operating-system/ thread? I can't help but notice striking similarities. In both cases, a guy somewhere is convinced that the big, dumb, greedy industry is pushing unnecessarily complex, proprietary and expensive solutions as a part of a conspiracy to extort more money from stupid engineers who don't realize they could do the same with a magical $0.10, 8-pin, 1GHz part. Isn't that funny that all the Dawn OS guy needed to market his revolutionary operating system was a non-existent, cheap, small and screamingly fast one-operand SUBLEQ CPU? I say we have to put those two in touch and let them both free our sorry bunch (and the rest of humanity) from the tyranny of proprietary, over-complicated, too-much-pins, too-many-instructions, oppressive technology? Unless they're the same guy, that is... 

[JPortici]

Anyone remember this: https://www.eevblog.com/forum/projects/dawn-operating-system/ thread? I can't help but notice striking similarities

my thoughts exactly

it'd be nice if more replies had at some arguments to support the claims

meanwhile: 

[MK14]

Unless they're the same guy, that is... 

I think they are completely different people.
The OP, does NOT remind me of Geri.

To me, they are NOTICEABLY different people. With completely different styles.

[Zbig]

Unless they're the same guy, that is... 

I think they are completely different people.
The OP, does NOT remind me of Geri.

To me, they are NOTICEABLY different people. With completely different styles.

Oh well, I guess you're 2 million percent RIGHT (just to pull some arbitrary large number out of thin air, for no particular reason).

[MK14]

Unless they're the same guy, that is... 

I think they are completely different people.
The OP, does NOT remind me of Geri.

To me, they are NOTICEABLY different people. With completely different styles.

Oh well, I guess you're 2 million percent RIGHT (just to pull some arbitrary large number out of thin air, for no particular reason).

Well I'm 2,000,001% sure that my gut feeling about Geri, being a different person, might be right. But it is just one persons opinion, which could also be wrong.

[ali_asadzadeh]

Use V3S from allwinner! it's 32Bit, 1GHz and have internal 64MB RAM, it's 128QFP, the only draw back is the poor documentation and linux resources

[rsjsouza]

Dave explored this subject a long time ago, but with FPGAs. He wanted to get a very large FPGA in a smaller package.

Short answer: there is no high scale market for such thing.

Long answer:
In my experience, the vast majority of applications of fast cores focus on the ability to perform multiple operations in "parallel" (true if multi core) or "fast sequenced" (if single core) while interacting with the real world in multiple ways - either via a plethora of serial interfaces or via very fast dedicated buses (USB2/3, Multi-lane SRIO, etc.) or external devices such as RAM. That, tied to the required additional pins power power and GND and the need to exhaust thermal energy fast enough from the device before it melts, increases the number of pins considerably. 

You could argue that you can use PowerPAD for thermal dissipation, reduce the number of peripherals and maybe even remove the external RAM controller. Thermally speaking this would offset the absence of pins on the package; however, how to decide which peripherals to remove? Would your application require only I²S (for audio) or fast SPI? What about the customer that would also need some I²C or a small 16-bit memory interface for parallel ADCs? Spin another device variant?

If you want to categorize the device as a "MCU", then it would be almost mandatory to add non-volatile memory inside the device and increase the cost sensibly (not to mention the higher thermal profile inside the package tends to reduce endurance sensibly). Running 1GHz from Flash is impossible, thus you would need a great amount of internal high speed RAM. If you want to remove the external high speed RAM interface, how much internal RAM would be enough, given it would have to accommodate data+code?

Therefore, to answer your question as to why there are no 1GHz MCUs: there is considerable risk and cost associated with the release of a family of devices and nobody will target a niche market (unless there are heaps of money to be made) or a hot market of yore (audio processing, as you provided as an example) where other dedicated solutions exist at a fraction of the cost of a general programmable device.

So yes, you are right: there are no "real" choice's, only expensive ""forced-upon-you" solutions. - the forces are beyond you, I, the manufacturers and are imposed by "the market". Obviously this can change anytime, provided the money follows it - one example is the Allwinner device mentioned above. If it is successful, other manufacturers will follow.

[bson]

I could see lots of interesting applications for a small 1GHz uC... It wouldn't need much memory, maybe 512 bytes, certainly less than a typical L1 cache.  It could probably execute very close to 1G instructions per second out of a very wide NOR flash, where a line is wide enough to encompass the typical branch range.  This would also permit a very simple core without branch prediction or speculative execution, read/write queues, etc.  Just a basic Cortex-M core.  Such a uC could do a lot of interesting tasks in hardware, for example a reasonably accurate phase detector or a software PLL, or many other handy little glue tasks - like timing generation with very precisely programmable skews.  It could also have very low interrupt latency.  If it has a fast ADC built in it can do even more.  I see no reason why not, but of course fast RAM for it will take up quite a bit of die space, but it wouldn't need much.

[brucehoult]

I could see lots of interesting applications for a small 1GHz uC... It wouldn't need much memory, maybe 512 bytes, certainly less than a typical L1 cache.  It could probably execute very close to 1G instructions per second out of a very wide NOR flash, where a line is wide enough to encompass the typical branch range.  This would also permit a very simple core without branch prediction or speculative execution, read/write queues, etc.  Just a basic Cortex-M core.  Such a uC could do a lot of interesting tasks in hardware, for example a reasonably accurate phase detector or a software PLL, or many other handy little glue tasks - like timing generation with very precisely programmable skews.  It could also have very low interrupt latency.  If it has a fast ADC built in it can do even more.  I see no reason why not, but of course fast RAM for it will take up quite a bit of die space, but it wouldn't need much.

For such, Altera has 3mm*3mm MAX10 FPGAs with 2k LEs and plenty of RAM. You can implement a NiosII and accelerate key logic with FPGA. It also has a few hard multiplier blocks and some of its RAM can be used as soft multipliers as well.
Sounds to be a cheap choice for me, considering the price tag (2x~10x common Cortex M0 MCUs) and the small side, as long as you have a way to deal with 0.4mm pitch BGA, it pretty much crashes MCUs in terms of digital performance. BTW, it comes with flash as well, so no EPCS needed.

Five or six euros for the 2000 PE version. Not bad.