On the feasibility of handling manufacturing faults in embedded memories by software means

This paper presents a software-based technique for handling permanent faults in low-cost memories for embedded systems. Basically the technique is based on an adaptation of the program code that avoids the usage of faulty memory words. With this technique it is possible to handle permanent manufacturing faults in simple and cheap memories without any kind of hardware-based fault management in the memory. This is of particular interest for long living low-cost embedded systems like sensor nodes that make use of such memories. Different methods for handling memory faults in data and program memory are discussed. The methods for program memories are evaluated with flash memory devices. Based on real test data for the flash memories it is shown that with the software-based technique a significant amount of hardware-overhead in the memory can be avoided without losing the capabilities of handling memory faults.