A memory device includes a memory array, a first buffer, a second buffer, a repair logic circuit and an internal memory. The method of operating the memory device includes: the repair logic circuit receiving a bad column table from the internal memory, the bad column table containing information of a bad column in the memory array; the first buffer receiving first data; the repair logic circuit receiving the first data from the first buffer; and the repair logic circuit mapping the first data onto second data according to the bad column table.

Many error correction schemes fail to correct for double-bit errors and a module must be replaced when these double-bit errors occur repeatedly at the same address. This helps prevent data corruption. In an embodiment, the addresses for one of the memory devices exhibiting a single-bit error (but not the other also exhibiting a single bit error) is transformed before the internal memory arrays are accessed. This has the effect of moving one of the error prone memory cells to a different external (to the module) address such that there is only one error prone bit that is accessed by the previously double-bit error prone address. Thus, a double-bit error at the original address is remapped into two correctable single-bit errors that are at different addresses.

To overcome a problem of increase of test time related to BIST in a conventional semiconductor device, a semiconductor device according to one embodiment includes a plurality of memory arrays having different sizes, a test pattern generation circuit that outputs a test pattern for the memory arrays, and a memory interface circuit that is provided for every memory array and converts an access address. The memory interface circuit shifts a test address output from the test pattern generation circuit in accordance with a shift amount set for every memory array, thereby converting the test address to an actual address of a memory array to be tested.

A semiconductor memory device includes: a first wiring and a second wiring; a first selection transistor, a memory transistor, and a second selection transistor connected between the first wiring and the second wiring; and a third wiring and a fourth wiring connected to gate electrodes of the first selection transistor and the second selection transistor. From a first timing to a second timing, a first voltage that turns the first selection transistor ON is supplied to the third wiring, and a second voltage that turns the second selection transistor OFF is supplied to the fourth wiring. From the second timing to a third timing, a third voltage that turns the first selection transistor OFF is supplied to the third wiring, and at a fourth timing between the first timing and the third timing, at least one of a voltage and a current of the first wiring is detected.

This disclosure relates to testing of integrated artificial intelligence (AI) circuit with embedded memory to improve effective chip yield and to mapping addressable memory segments of the embedded memory to multilayer AI networks at the network level, layer level, parameter level, and bit level based on bit error rate (BER) of the addressable memory segments. The disclosed methods and systems allows for deployment of one or more multilayer AI networks in an AI circuit with sufficient model accuracy even when the embedded memory has an overall BER higher than a preferred overall threshold.

According to one embodiment, a memory system manages a plurality of parallel units each including blocks belonging to different nonvolatile memory dies. When receiving from a host a write request designating a third address to identify first data to be written, the memory system selects one block from undefective blocks included in one parallel unit as a write destination block by referring to defect information, determines a write destination location in the selected block, and writes the first data to the write destination location. The memory system notifies the host of a first physical address indicative of both of the selected block and the write destination location, and the third address.

A semiconductor device includes a mode setting circuit configured to allocate any one of values to a mode signal based on an event signal, an address converter configured to generate a conversion address by converting at least one address based on the mode signal, and a memory circuit configured to perform an operation corresponding to the conversion address.

A processor memory is stress tested with a variable link stack depth using test code segments and link stack test segments on non-naturally aligned data boundaries. Link stack test segments are interspersed into test code segments of a processor memory test to change the link stack depth without changing results of the test code. The link stack test segments include branch to target, push/pop, push and pop segments. The depth of the link stack is varied independent of the memory test code by changing the number to branches in the branch to target segment and varying the number of the push/pop segments. The link stack test segments and test segments may be placed randomly with a recursive algorithm to intersperse the link stack test segments in the test code segments and to reduce the amount of data to be saved and restored for all subroutine calls, push and pop segments.

This disclosure relates to testing of integrated artificial intelligence (AI) circuit with embedded memory to improve effective chip yield and to mapping addressable memory segments of the embedded memory to multilayer AI networks at the network level, layer level, parameter level, and bit level based on bit error rate (BER) of the addressable memory segments. The disclosed methods and systems allows for deployment of one or more multilayer AI networks in an AI circuit with sufficient model accuracy even when the embedded memory has an overall BER higher than a preferred overall threshold.

A semiconductor memory device includes a memory cell array, a read/write circuit, and a control logic. The memory cell array includes a plurality of memory blocks. The read/write circuit performs a read/write operation on a selected page of the memory cell array. The address decoder stores bad block marking data on each of the plurality of memory blocks, and outputs the bad block marking data in response to an address signal. The control logic controls the read/write circuit to test whether a defect has occurred in the plurality of memory blocks, and controls the address decoder to store, as the bad block marking data, a test result representing whether the defect has occurred in the plurality of memory blocks.