If not relying on an overly strict definition of "data memory", Andy's statement is in fact completely true. Keil's XDATA banking feature can be exploited to provide a relatively nice abstraction that an I²C-attached serial EEPROM is actually ordinary read/write memory from the perspective of a C application. Details may be found here:

http://www.8052.com/forum/read.phtml?id=24473

Of course, no amount of clever abstraction changes the fact that a serial EEPROM is not a suitable substitute for a general purpose data memory: although the serial EEPROM carries the benefit of nonvolatility, the I²C interface of the serial interface will make it slow compared to parallel devices, and the fact that it is an EEPROM fundamentally limits its write performance and maximum write cycle count. An SRAM device, on the other hand, would not be subject to these limitations.


Outside of being correct, Roberto alludes to an important subtlety here: EEPROM devices, in most cases, can be conveniently written/rewritten in the target without specialized hardware. EPROM devices, on the other hand, typically require sepcialized hardware to write (e.g., hardware not normally present in an embedded target), and if they can be erased at all, typically require exposure to a UV light source. Despite seemingly similar part numbers, the '24C64 and '27C64 are in fact radically different devices.

Note that the '28C64, an EEPROM version of the '27C64, can readily be von Neumann wired (e.g., mapped into both CODE and XDATA address spaces) to serve as a bootable program device while being simultaneously in-application readable and writable. The main motivations for such a configuration would typically be in-application code updating or persistent parameter storage.

--Sasha Jevtic