In my application, I have one device always connected to the SPI Bus where the McBSP is the Master and an AT45DB161D Flash Memory chip is the SPI slave. This device is fairly complex and supports many commands. They all work based on a command followed by a response. To send a command, a message (in my case 8 bytes long) and then a response occurs with data transmitted back for as long as your clock it. During the entire time, the memory chip select must remain active. The command I send is a Read Command of a page of Flash Memory.
I am reading the data from a series of Flash Memory Pages which happen to be 528 Bytes long. Due to the limitations of the TMS320C6713B McBSP port namely that the chip select can not be programmed to be active for more than 4096 clock cycles, I broke-up the problem of reading the flash page into 8 chunks each 74 bytes long with 8 bytes associated with the command and 66 bytes of data.
I programmed the RCR and the XCR as single phase system with a length set to 73 decimal (elements) with each element set to 8 bit length to give me a 74 byte message.
I have the EDMA channels associated with the McBSP transmitter and receiver set-up to transmit the desired command followed by zero data in the case of the transmitter and a receiver buffer set-up to receive the data from the Flash Memory chip.
We have been procuring boards from a third party since 2008. They use the a 300MHz DSP chip in the 272 Pin PGA package. We take the board and derate it down to run a 200 MHz and run the external bus a 80 MHz. We have used 72 boards with all but 3 boards operating down to -40 C. Those 3 boards operate at least to -20 C and then problems appear to happen. At the coldest temperature, the chip select (i.e., the FSX output) is eratic (not a logic 1 or zero) but a signal not well synchroized to the clock and could be at a low voltage less than a volt. It never returns to a high level.
The processor itself is fully operation. Only the McBSP appears affected. Could you find out if the chip has known performance issues relating to the McBSP over temperature and when parts screened to -40 C to 125 C this capability is tested. We are concerned because when the last 10 boards were procured from our supplier, 2 boards in the group failed at cold temperature.
The Clock appears to look the same on a scope, occasionally the DX output transitions to low and high a little slow with the signal below 0.5 volts taking much longer to decay. As stated before the FX output looks really bad (not able to go to a legal "1" voltage.
I have tried to run the McBSP at 5 MHz vs 10 Mhz and it has no affect on the temperature behavior.
Before the signal degradation is observable on a scope, the memory chip fails to respond. I am always able to read the transmitted data with my Total Phase Analyzer in all cases. The decay of the DX line and the rise only occurs near the edges of the FX transitions well before any clock is activated (which explains why the Beagle Analyzer is unaffected by the degration).
We have tested with a few I/O Boards and a few CPU Boards, the problem always follows the CPU Boards.
Bob Bianchi
