Fast simultaneous sampling for reasonable price

[poorchava]

W're beggining a project, which needs to sample 3 signals simultaneously quite fast (~500k-1Msps) at resolution higher than standard 12-bits found in most microcontrollers. Start of sampling must be really simultaneous. <10ns difference is desired. We also need to process that data which involves standard dsp stuff (fft and such). In the previous version of the product we used a 120Mhz CM4 microcontroller and it has proven to be significantly too slow for what we anted to do. We estimate the the new version of the algorithm will require about 500k-1M of RAM. We'd ideally like the design to be implementable on a standard 4-layer PCB. 6-layer is a possibility if we are absolutely forced to, but more layers, HDI or other exotic processes are unacceptable.

Now it seems like the new STM32H7 family from ST fits the bill perfectly. Lots of flash, whopping 1MB of SRAM, 850DMIPS/400MHz, 3 separate 16-bit ADCs with hardware oversampling and differential input capability (although even in diff mode ENOB is ~13.5). There are two problems with that: first of all it doesn't seem like H7s will be available in retail any time soon. We are currently trying to get a couple of devboards or even bare parts themselves through our suppliers, but it doesn't look promising. Another thing is that H7 has shitload of stuff we don't need and ideally wouldn't want to pay for (we only need the ADC, 10 mbit serial port and shitload of processing power, ok maybe an external memory controler will come in handy too).

Can you think of another solution to that problem other than a microcontroller with fast sampling integrated ADCs? We've looked into solutions like standalone s/h chips, implementing a discrete s/h using multiplexers, external adc with parallel sampling functionality, parallel bus adcs, fpgas and such, but anything we come up with immediately turns out to be far more expensive than the H7 approach. Implementing entire thing in an FPGA is gonna require massively expensive HW (it onvolves 4k point F32 ffts) and loads of development time. The only somewhat reasonable solution that came up is to use 3 independent ADCs hooked up to a cheap FPGA which then connects to a powerful uC over SRAM interface, but it still turns out to be more expensive than an H7 and obviously more complicated than the H7 variant from software development standpoint.


Perhaps there is some obscure family of microcontrollers that would fit the bill?