Microprocessor memory is sensitive to cold boot attacks. In this kind of attacks, memory remanence is exploited to download its content after the microprocessor has been struck by a hard boot. If just in this moment, a crypto-algorithm was in execution, the memory data can be downloaded into a backup memory and specialized tools can be used to extract the secret keys. In the main memory data can be protected using efficient encryption techniques but in caches this is not possible unless the performance becomes seriously degraded. Recently, an interleaved scrambling technique (IST) was presented to improve the security of caches against cold boot attacks. While IST is effective for this particular kind of attacks, a weakness exists against side channel attacks, in particular using power analysis. Reliability of data in caches is warranted by means of error detecting and correcting codes. In this work it is shown how these kinds of codes can be used not only to improve reliability but also the security of data. In particular, a self-healing technique is selected to make the IST technique robust against side channel attacks using power analysis.