Memory that is accessed according to the characteristics of a given content rather than by address.The characteristics of associative memory are: ① In addition to storage function, it also hasinformation processingFunction.It can search according to the characteristics of the content sentStorage unit。②For eachStorage unitThe parallel search can significantly improve the search speed.These characteristics are similar to the "associative" function of the human brain, so they are called associative memory.[1]
Stored in memoryinformationIt can be described by two parameters: address and numerical value (content).If twoStorage unitIf the content in is used for some operation and the result is stored in one of the cells, the memory accessed by address is more suitable.If you search based on certain content characteristicsStorage unit, the result can be obtained faster by using associative memory.andstorage space The usage of is relatively economical, and the cost of modifying recordsSystem overheadIt is also small.Associative memory is used in largedata baseReal-time retrieval and update of, andGeneral purpose computerOfVirtual StorageAnd control system.[1]
form
Announce
edit
The figure shows the composition of associative memory.EveryStorage unitBoth contain circuits for storage, comparison, reading and writing, and control.The search variable is stored in the comparison numberregisterMedium.shieldregisterIt is used to mask some contents of the comparison number register, while the unmasked part is sent to the associative memory as a lookup variable.When searching, eachStorage unitCompare its contents with the sent lookup variable.If equal, respondregisterThe corresponding bit in is set to 1.If not, set 0.In this way, the search results will be stored in the responseregisterMedium.Satisfying the search requirementsStorage unitIt is called response unit.The search is performed in parallel, so there may be more than one response unit, which is called multiple responses.At this time, if you want to writeinformation, the information can be sent to the comparison number firstregister, and then sent to Lenovo memory through cooperation with the mask register, and written to these response units in parallel.Sometimes it can be written according to the address of the unitinformation。To readinformation, these cells must be decomposed one by one to determine the address order for reading, which is called multiple response decomposition and is completed by the multiple response resolver.Word selectionregisterUsed to select theStorage unit, plays the role of inter word shielding."Comparison" is the most basic logical operation of associative memory, which can complete a variety of comparison operations for a given search variable.For example: all, unequal;Less than, greater than;Not greater than, not less than;Only greater than, only less than;In section, out of section, maximum value, minimum value, etc.These operations can be realized through corresponding algorithms on the basis of full comparison, or inStorage unitAdd the corresponding logic circuit to undertake.[1]
Insufficient
Announce
edit
Although associative memory has strong functions, it contains many circuits, so its cost is very high.Computer systems are often configured with such memory in a small scale to improve the performance price ratio of the whole system.[1]
Application in paging system
Announce
edit
In a paging system, the CPU accesses the memory twice every time it accesses a data (accessing the page table and the actual physical address).In order to improve the speed of address transformation, a special cache with parallel query capability, called "associative memory" or "fast table", is added to store the currently accessed page table entries.[1]
Bidirectional associative memory
Announce
edit
In the bidirectional associative memory model, a set of finite vector pairs establish the bidirectional associative relationship by a linear operator, and this linear calculation is the connection weight matrix of a network.The weight matrix is determined by the least square method.A special type of Lyapunov matrix equation is derived from the solution of the weight matrix.[1]
Bibliography
Announce
edit
Jin Lan, Wang Dingxing, Shen Meiming, Parallel Processing Computer Architecture, National Defense Industry Press, Beijing, 1982.[1]