Associative memory

from Wikipedia, the free encyclopedia

An associative memory or content addressable memory (Engl. Content Addressable Memory , CAM) is a memory shape, in which the association is working content, to access individual memory contents. Another paraphrase would be that access to a memory content takes place by entering a memory value and not via a memory address .

The special nature of an associative memory is based on the structure organization that is independent of the respective access - best understandable with the question: 'What is in memory locations that have not yet been written?'. While addressing with explicit addresses are regularly a list or table structure again, in which the memory locations defined it first with an empty value are occupied, arise in idealized associative memory only there storage locations for the above, a write operation has taken place.

history

The associative memory was proposed by Konrad Zuse as early as 1943 , but could only be implemented with the development of semiconductor technology . Today, associative memory has been largely replaced by hashing techniques that work with conventional memory and are therefore much cheaper to implement.

Areas of application

Fast cache memories are a typical area of ​​application . Memory tables that are frequently accessed can also be implemented as associative memories. For example, in high-quality network switches, the table of MAC addresses is designed as a CAM. Another area of ​​application is data structures in artificial intelligence . This is how human memory works with associations ; z. B. it associates memories of certain experiences with certain objects.

It is difficult for computer systems to implement large associative memories, since “fuzzy” inputs should also lead to a desired result ( fuzzy search ). In general, an input vector should provide an output vector if it has enough similarity to a pattern vector . Neural networks are used to implement such functions . A simple implementation of the described assignment is possible, for example, through one of the simplest neural networks, the single-layer perceptron .

Special forms

A Ternary Content Addressable Memory  (TCAM) works with a third logical value , Don't Care , in order to access the memory contents.

If several associative memories are combined in such a way that, in addition to data, programs can also be stored and executed, an associative machine is created .

Individual evidence

  1. cisco.com: CAM VS TCAM