diff options
Diffstat (limited to 'notes/fsmc.txt')
| -rw-r--r-- | notes/fsmc.txt | 51 |
1 files changed, 36 insertions, 15 deletions
diff --git a/notes/fsmc.txt b/notes/fsmc.txt index 583dba2..b41de60 100644 --- a/notes/fsmc.txt +++ b/notes/fsmc.txt @@ -1,18 +1,32 @@ -FSMC notes (for maple native) +FSMC notes (for maple native and other "high density" STM32 devices) ------------------------------------------------------------------------------- There is an application note for all this which is helpful; see the ST website. -Chip details +SRAM chip details IS62WV51216BLL 512k x 16 19 address input 16 data inputs + t_wc (write cycle) = 55ns + t_rc (write cycle) = 55ns + t_pwe1 (write enable low pulse) = 40ns + t_aa (address access) = 55ns -For simple debugging, i'm going to set all the access parameters to maximum -time values (aka, slowest). I'm going to use not-extended mode 1 for -read/write. + +The FSMC nomenclature is very confusing. There are three seperate "banks" +(which I will call "peripheral banks") each of specialized for different types +of external memory (NOR flash, NAND flash, SRAM, etc). We use the one for +"PSRAM" with our SRAM chip; it's bank #1. The SRAM peripheral bank is further +split into 4 "banks" (which I will call "channels") to support multiple +external devices with chip select pins. I think what's going on is that there +are 4 hardware peripherals and many sections of RAM; the docs are confusing +about what's a "block of memeory" and what's an "FSMC block". + +Anyways, this all takes place on the AHB memory bus. + +I'm going to use not-extended mode 1 for read/write. Steps from application note: @@ -22,21 +36,28 @@ Steps from application note: - memory is nonmultiplexed: BCR3_MEXEN is reset (= '0') - everything else is cleared -Parameters: - - t_wc (write cycle) = 55ns - t_rc (write cycle) = 55ns - t_pwe1 (write enable low pulse) = 40ns - t_aa (address access) = 55ns +But not true! Actually write enable needs to be set. -So address setup (ADDSET) = 0x0, data setup (DATAST) = 0x3 +Using the application note, which is based around a very similar chip (with +faster timing), I calculated an ADDSET (address setup) value of 0x0 and a +DATAST (data setup) value of 0x3. -Using bank1, NOR/PSRAM1 memory starts at 0x60000000. +Using channel1, NOR/PSRAM1 memory starts at 0x60000000. -Oops, obviously have to turn on the clock for all those GPIO pins... +Have to turn on the RCC clock for all those GPIO pins, but don't need to use +any interrupts. Not-super-helpful-link: http://www.keil.com/support/man/docs/mcbstm32e/mcbstm32e_to_xmemory.htm -PG9 (which is NE2) is twiddling on reset? +Note the possible confusion with address spaces, bitwidths, rollovers, etc. + + +TODO +------------------------------------------------------------------------------- +- more rigorous testing: throughput, latency, bounds checking, bitwidth, data + resiliance, etc. +- update .ld scripts to transparently make use of this external memory +- test/demo using a seperate external SRAM chip or screen +- write up documentation |