A Method for Perceiving Threshold Voltage Distribution of 3D NAND Flash Memory Block Granularity

Abstract

The present invention proposes a method for perceiving the threshold voltage distribution at the block granularity of 3D NAND flash memory based on a tag layer. First, the tag layer in the block is obtained offline at the initial stage of data storage and recorded in the storage metadata area. During the data storage cycle, the threshold voltage distribution at the block granularity can be quickly obtained by perceiving the tag layer, and then reliability management parameters based on the threshold voltage distribution, such as optimized read reference voltages, can be obtained. Through in-depth characterization of the inter-layer differences in the threshold voltage distribution of flash memory and using the proposed tag layer method, the present invention significantly reduces the management overhead for obtaining the threshold voltage distribution at the block granularity and improves the reading performance. It realizes the fast perception of the optimized read reference voltage at the block granularity and can be adapted to many existing flash memory reliability management strategies based on the optimization of read reference voltages.

Description

Technical Field

[0001] The present invention belongs to a storage system of a flash memory medium, and specifically, relates to a method for sensing a threshold voltage distribution at the block granularity of a 3D NAND flash memory based on a marking layer. Background Art

[0002] In currently mainstream 3D flash memories on the market, a block is composed of several stacked layers. The internal circuit of the flash memory realizes non-volatile storage by injecting charges into cells. After the charges are injected into the cells, a threshold voltage is formed thereon. The more charges are injected, the higher the formed threshold voltage value. The programming process is to inject an appropriate amount of charges into the cells to raise their threshold voltages to the states corresponding to the stored information. Different threshold voltage states correspond to different stored information. Taking the currently widely used 3-bit / cell (triple-level cell, TLC) flash memory as an example, its cells can be set to eight states including ER, A, ……, G for realizing the storage of 3-bit information. During the reading process, the internal circuit of the flash memory determines the threshold voltage of the cell by reading the reference voltage, and then identifies the state to which it is programmed (that is, obtains the 3-bit information corresponding to this state). When the read reference voltage applied to the cell is less than its threshold voltage, the cell is not conductive. On the contrary, when the read reference voltage applied to the cell is greater than its threshold voltage, the cell is conductive. The critical read reference voltage that just makes the cell conductive is recognized by the flash memory circuit as the threshold voltage of the cell. It should be noted that due to process differences, even at the moment when programming has just ended, there are still differences in the threshold voltages of cells programmed to the same state, manifested as a distribution within a certain range, which is called the threshold voltage distribution. As Figure 1 shown, it is the correspondence between the threshold voltage and the stored 3-bit information of a 3D TLC flash memory chip. For example, if the information to be stored in a certain cell is "101", then during the programming process, the internal circuit of the flash memory will inject charges into it to make it in the "C" state. During the reading process, the internal circuit of the flash memory first identifies that the threshold voltage range of this cell is within [V c , V d , and then identifies the 3-bit information "101" corresponding to the C state.

[0003] In an ideal state, the amount of charge captured by a cell during the data storage cycle remains unchanged, which means that the threshold voltage of the cell also does not change. In actual situations, reliability stresses such as retention loss will cause the amount of charge captured by the cell to vary, thereby causing the threshold voltage of the cell to change. Macroscopically, this will lead to a drift in the threshold voltage distribution of the state. Specifically, if a cell leaks charge during data storage, resulting in the threshold voltage distribution exceeding the left boundary of the original state window, then a downward shift error will occur, and these cells will be recognized as the left adjacent state of the original state window during the read operation. Conversely, if a cell is injected with charge during data storage, an upward shift error will occur and it will be recognized as the right adjacent state of the original state window.

[0004] As can be seen from the above, data stored in the flash memory medium will experience errors. With the advancement of flash memory integration technology, the cost of increasing storage density is the deterioration of storage reliability. Therefore, current mainstream flash memory device manufacturers will all equip appropriate reliability management solutions in the flash memory controller. Obtaining the threshold voltage distribution at the block granularity is a core part of many flash memory reliability management solutions. Taking the application of optimizing the read reference voltage as an example. Current mainstream 3D flash memory technology supports modifying the read reference voltage and performing read operations with the optimized voltage. As Figure 2 shown, if reading is performed with the default read reference voltage (V default ), more error codes will be generated, while reading with the optimized read reference voltage (V OPT ) can significantly reduce the error codes. Existing methods often need to perform multiple rounds of read operations on the entire block to obtain the threshold voltage distribution at the block granularity in order to obtain the optimized reference voltage at the block granularity, which often brings longer read latency and management overhead, imposing a large burden on system performance in read-intensive storage scenarios. Therefore, how to quickly and accurately obtain the threshold voltage distribution at the block granularity is an important challenge faced by many flash memory reliability management solutions. Summary of the Invention

[0005] Based on the above problems, the present invention proposes a method for perceiving the threshold voltage distribution at the block granularity of a 3D NAND flash memory based on a marker layer. First, the marker layer in the block is obtained offline at the initial stage of data storage and recorded in the storage metadata area. During the data storage cycle, the threshold voltage distribution at the block granularity can be quickly obtained by perceiving the marker layer, and then reliability management parameters based on the threshold voltage distribution, such as the optimized read reference voltage, can be obtained.

[0006] The present invention is realized through the following technical solutions:

[0007] A method for perceiving the threshold voltage distribution at the block granularity of a 3D NAND flash memory based on a marker layer:

[0008] The method specifically includes the following steps:

[0009] Step 1: First, obtain the mark layer of each adjacent state pair in the block granularity offline at the initial stage of data storage, and record it in the storage metadata area;

[0010] Step 2: During the data storage cycle, the threshold voltage distribution of the block granularity is quickly obtained by sensing the mark layer;

[0011] Step 3: Obtain the optimized read reference voltage based on the threshold voltage distribution of the block granularity.

[0012] The first step specifically includes the following steps:

[0013] Step 1.1: Write data and apply residence loss stress to allow the stored data to overcome the early unstable rapid residence loss stage and exhibit a long-term residence loss error mode;

[0014] Step 1.2: Obtain the distribution of the target to the adjacent states layer by layer;

[0015] Step 1.3: Get the distribution of the target to adjacent states of the entire block and divide it by the number of layers to get the normalized block granularity distribution;

[0016] Step 1.4, calculate the root mean square error (RMSE) between the layer particle size distribution and the normalized block particle size distribution layer by layer;

[0017] Step 1.5: Retrieve the minimum value in the RMSE sequence obtained in step 1.4 and record its corresponding layer number as the marker layer of the adjacent state of the target pair;

[0018] Step 1.6: Record the marking layer serial number obtained in step 1.5 into the metadata area of the flash memory for later use in the data storage cycle.

[0019] Furthermore, in step 1,

[0020] Assume that the adjacent state pairs to be obtained are X state and Y state respectively.

[0021] For layer numbered i, the distributions of X state and Y state are f Xi (ν), and f Yi (ν),

[0022] For the entire block, the distribution of X state and Y state are g Xi (v), and g Yi (v)

[0023] The actual distribution is normalized by the number of layers to obtain the block size distribution.

[0024] The root mean square error is used to measure the distribution difference between a layer and a block. The calculation formula is:

[0025]

[0026] where N is the number of sampling points of the X and Y states;

[0027] The marked layer is the layer that satisfies the minimum RMSE. For adjacent X and Y states, the distribution of the marked layer is closest to the normalized distribution of the entire block.

[0028] Further, in step two,

[0029] After performing step 1.6 twice, the read reference voltage optimized by the predicted block granularity of the marked layer is used for reading to reduce the number of error bits.

[0030] A 3D NAND flash block granularity threshold voltage distribution sensing device based on a marked layer:

[0031] The device includes: an offline module, an online module, and a voltage management module

[0032] The offline module is used to offline obtain the marked layer of each adjacent state pair in the block granularity at the initial stage of data storage and record it in the storage metadata area;

[0033] The online module is used to quickly obtain the threshold voltage distribution of the block granularity by sensing the marked layer during the data storage period;

[0034] The voltage management module is used to obtain the optimized read reference voltage according to the threshold voltage distribution of the block granularity.

[0035] Further, the offline module further includes:

[0036] The data writing module is used to write data and apply a retention loss stress to let the stored data cross the early unstable fast retention loss stage and exhibit a long-term retention loss error pattern;

[0037] The state acquisition module is used to layer by layer obtain the distribution of the target pair of adjacent states;

[0038] The normalization processing module is used to obtain the distribution of the target pair of adjacent states of the entire block and divide it by the number of layers to obtain the normalized block granularity distribution;

[0039] The root mean square error calculation module calculates the root mean square error RMSE between the layer granularity distribution and the normalized block granularity distribution layer by layer;

[0040] The retrieval module is used to retrieve the minimum value in the RMSE sequence obtained by the root mean square error calculation module and record the corresponding layer number as the marked layer of the target pair of adjacent states;

[0041] The recording module is used to record the obtained marker layer numbers in the retrieval module into the metadata area of the flash memory for later calls during the data storage cycle.

[0042] Further, let the adjacent state pairs to be obtained be the X state and the Y state respectively.

[0043] For the layer numbered i, the distributions of the X state and the Y state are f Xi (ν), and f Yi (ν),

[0044] For the entire block, the distributions of the X state and the Y state are g Xi (v), and g Yi (v),

[0045] Then, after normalizing the actual distribution by dividing it by the number of layers, the block-level distribution is obtained.

[0046] An electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0047] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps of the above method are implemented.

[0048] Advantages of the present invention

[0049] The present invention deeply characterizes the difference in the threshold voltage distribution between flash memory layers. Through the proposed marker layer method, the management overhead for obtaining the block-level threshold voltage distribution can be significantly reduced, and the read performance can be improved.

[0050] The present invention does not require excessive additional storage overhead, and only needs to record the physical address of the marker layer of the flash memory block in the metadata area.

[0051] The present invention uses the root mean square error to measure the difference in the threshold voltage distribution between the layer and the block (normalized), and selects the layer with the minimum root mean square error measure as the marker layer; adopts a marker layer-based method to quickly perceive the threshold voltage distribution at the block level; and can quickly obtain the optimized read reference voltage at the block level, realizing the fast perception of the optimized read reference voltage at the block level online, and can be adapted to many existing flash memory reliability management strategies based on the optimization of the read reference voltage. Description of the drawings

[0052] Figure 1 It is a schematic diagram of the threshold voltage distribution of the existing 3D TLC type NAND flash memory;

[0053] Figure 2 It is to reduce the read data error rate by modifying the read reference voltage between adjacent states;

[0054] Figure 3 Schematic diagrams of a pair of adjacent state marking layers, where (a) is the distribution diagram of all layers and the entire block, and (b) is the distribution diagram of the marking layer, the comparison layer, and the entire block;

[0055] Figure 4 Flow chart for obtaining a pair of target adjacent state marking layers;

[0056] Figure 5 Deviation between the layer distribution and the block distribution (normalized) measured by RMSE layer by layer in two retention cases;

[0057] Figure 6 Comparison of the threshold voltage distributions of the block (normalized), the marking layer, and the comparison layer in two retention cases, where (a) is for 4 months and (b) is for 6 months;

[0058] Figure 7 The optimized reference voltage obtained through the marking layer achieves the effect of reducing the number of error bits. Specific implementation manners

[0059] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0060] The threshold voltage distribution of the entire flash memory block is the average effect of the layer distributions. Currently, mainstream 3D NAND flash memories usually have very dense stacked layer numbers, and the process differences are mainly reflected between layers. As a physical whole, the inside of a layer has a similar distribution. Therefore, it is possible that a certain layer has a distribution very similar to that of the entire block. As Figure 3 shown in (a), the blue line is the distribution of each layer, and the red line is the distribution of the entire block (normalized by dividing by the number of layers). It can be seen that although there are certain differences in the distributions between layers, the continuity of this difference makes the distribution of a certain layer very close to the average effect of the distributions of all layers, that is, the normalized block distribution.

[0061] The above-mentioned difference in layer distributions is continuous during data retention, which is attributed to the stable process differences during data retention. For example, a layer with a large shift in the distribution mean has a large shift throughout the data storage period, and a layer with a large distribution variance also shows a relatively larger variance throughout the data retention period.

[0062] Combined with Figures 1 to 7 , the present invention proposes a method for perceiving the threshold voltage distribution of 3D NAND flash memory block granularity based on a marking layer:

[0063] The method specifically comprises the following steps:

[0064] Step 1: First, obtain the mark layer of each adjacent state pair in the block granularity offline at the initial stage of data storage, and record it in the storage metadata area;

[0065] Step 2: During the data storage cycle, the threshold voltage distribution of the block granularity is quickly obtained by sensing the mark layer;

[0066] Step 3: Obtain reliability management parameters based on threshold voltage distribution, such as optimized read reference voltage, according to the threshold voltage distribution of the block granularity.

[0067] The first step specifically includes the following steps:

[0068] Step 1.1: Write data and apply a certain amount of residence loss stress to allow the stored data to overcome the early unstable rapid residence loss stage and exhibit a long-term residence loss error mode;

[0069] Step 1.2: Obtain the distribution of the target to the adjacent states layer by layer;

[0070] Step 1.3: Get the distribution of the target to adjacent states of the entire block and divide it by the number of layers to get the normalized block granularity distribution;

[0071] Step 1.4, calculate the root mean square error (RMSE) between the layer particle size distribution and the normalized block particle size distribution layer by layer;

[0072] Step 1.5: Retrieve the minimum value in the RMSE sequence obtained in step 1.4 and record its corresponding layer number as the marker layer of the adjacent state of the target pair;

[0073] Step 1.6: Record the marking layer serial number obtained in step 1.5 into the metadata area of the flash memory for later use in the data storage cycle.

[0074] The specific method of offline obtaining the marking layer of each adjacent state pair in the block is:

[0075] Assume that the adjacent state pairs to be obtained are X state and Y state respectively.

[0076] For layer numbered i, the distributions of X state and Y state are f Xi (ν), and f Yi (ν),

[0077] For the entire block, the distribution of X state and Y state are g Xi (v), and g Yi (v)

[0078] The block granularity distribution is obtained by normalizing the actual distribution by the number of layers.

[0079] The distribution difference between a certain layer and the block is measured by the root mean squared error (RMSE), and the calculation formula is:

[0080]

[0081] where N is the number of sampling points of the X and Y states;

[0082] According to the above definition of the marked layer, the marked layer is the layer that satisfies the minimum value of RMSE. For adjacent X and Y states, the distribution of the marked layer is closest to the normalized distribution of the entire block.

[0083] In step two,

[0084] After executing step 1.6 twice, the optimized read reference voltage predicted by the marked layer is used for reading to reduce the number of error bits.

[0085] A 3D NAND flash block granularity threshold voltage distribution sensing device based on a marked layer:

[0086] The device includes: an offline module, an online module, and a voltage management module

[0087] The offline module is used to offline obtain the marked layer of each adjacent state pair in the block granularity at the initial stage of data storage and record it in the storage metadata area;

[0088] The online module is used to quickly obtain the threshold voltage distribution of the block granularity by sensing the marked layer during the data storage period;

[0089] The voltage management module is used to obtain the optimized read reference voltage according to the threshold voltage distribution of the block granularity.

[0090] The offline module further includes:

[0091] The reading module is used to offline obtain the marked layer of each adjacent state pair in the block granularity at the initial stage of data storage and record it in the storage metadata area;

[0092] The storage module is used to quickly obtain the threshold voltage distribution of the block granularity by sensing the marked layer during the data storage period;

[0093] The voltage management module is used to obtain reliability management parameters based on the threshold voltage distribution, such as the optimized read reference voltage, according to the threshold voltage distribution of the block granularity.

[0094] The reading module further includes:

[0095] The data writing module is used to write data and apply a retention loss stress to allow the stored data to cross the early unstable rapid retention loss stage and exhibit a long-term retention loss error pattern;

[0096] The state acquisition module is used to layer-by-layer acquire the distribution of the target adjacent states;

[0097] The normalization processing module is used to acquire the distribution of the target adjacent states of the entire block and divide it by the number of layers to obtain a normalized block granularity distribution;

[0098] The root mean square error calculation module calculates the root mean square error RMSE between the layer granularity distribution and the normalized block granularity distribution layer by layer;

[0099] The retrieval module is used to retrieve the minimum value in the RMSE sequence obtained by the root mean square error calculation module and record the corresponding layer number as the marked layer of the target adjacent states;

[0100] The recording module is used to record the marked layer serial number obtained in the retrieval module into the metadata area of the flash memory for later call during the data storage cycle.

[0101] Suppose the adjacent state pairs to be acquired are the X state and the Y state respectively,

[0102] For the layer numbered i, the distributions of the X state and the Y state are f Xi (ν), and f Yi (ν),

[0103] For the entire block, the distributions of the X state and the Y state are g Xi (v), and g Yi (v),

[0104] Then, the actual distribution is divided by the number of layers for normalization to obtain the block granularity distribution.

[0105] Example:

[0106] A set of test results for obtaining the marked layer and obtaining an optimized read reference voltage based on the marked layer to improve storage reliability. The test steps are as follows:

[0107] (1) Select an ordinary block and perform 5000 write / erase operations to simulate the wear effect of the flash memory chip in a real storage scenario;

[0108] (2) Write random data into this block;

[0109] (3) Place this chip in an environment of 85 °C for 4 hours to generate a retention loss equivalent to 2 months at room temperature;

[0110] (4) Scan the distribution of states A - G on each layer, calculate the RMSE of each pair of adjacent states layer by layer according to Equation (1), and then obtain the marked layer of each pair of adjacent states, where N is 64, that is, 32 sample points are taken for each state;

[0111] (5) Obtain the optimized read reference voltage between each pair of adjacent states in the marked layer;

[0112] (6) Read the data using the default read reference voltage and the optimized read reference voltage obtained in step (5) respectively, and obtain the bit error rates in the two reading cases;

[0113] (7) Then place the chip in an environment of 85 °C for 4 hours to generate a retention loss equivalent to 2 months at room temperature, that is, the stored data has experienced a retention loss equivalent to 4 months at room temperature;

[0114] (8) Execute steps (4) - (6) again in a round, and obtain the corresponding test results.

[0115] Taking the pair of adjacent states of state B and state C as an example, after executing step (4) twice, Figure 5 The RMSE between the distributions of state B and state C on each layer and the normalized distribution of the whole block in the two retention stages is shown. Other pairs of adjacent states (such as state E and state F) have similar test results. Generally speaking, the layer with a relatively smaller RMSE under the condition of 4 - month retention still shows a relatively smaller RMSE under the condition of 8 - month retention, indicating the stability of the RMSE difference between layers during the retention process. Among them, layer No. 41 is selected as the marked layer in this test, and layer No. 1 is selected as the comparison layer.

[0116] Furthermore, according to Figure 3 the principle shown, Figure 6 the threshold voltage distribution of the block (normalized), the marked layer, and the comparison layer in the two retention stages is obtained. It can be seen that under the conditions of 4 - month retention and 8 - month retention, layer No. 41, the marked layer, has a distribution very similar to that of the whole block, while the distribution of layer No. 1, the comparison layer, is quite different from that of the whole block.

[0117] After executing step (6) twice, reading using the optimized read reference voltage predicted by the marked layer for the block granularity, the effect of reducing the number of bit errors is as Figure 7 shown. For the stages of 4 - month retention and 8 - month retention, the bit errors of the whole block are reduced by 83% and 81% respectively, indicating that the optimized read reference voltage obtained based on the marked layer can effectively improve the reliability of stored data.

[0118] An electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0119] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps of the above method are implemented.

[0120] The above has introduced in detail a method for perceiving the threshold voltage distribution of a 3D NAND flash memory block granularity based on a marking layer proposed by the present invention, and has elaborated on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for sensing the block-size threshold voltage distribution of a 3D NAND flash memory based on a marker layer, characterized by: The method specifically comprises the following steps: Step 1: First, obtain the mark layer of each adjacent state pair in the block granularity offline at the initial stage of data storage, and record it in the storage metadata area; The first step specifically includes the following steps: Step 1.1: Write data and apply residence loss stress to allow the stored data to overcome the early unstable rapid residence loss stage and exhibit a long-term residence loss error mode; Step 1.2: Obtain the distribution of the target to the adjacent states layer by layer; Step 1.3: Get the distribution of the target to adjacent states of the entire block and divide it by the number of layers to get the normalized block granularity distribution; Step 1.4, calculate the root mean square error (RMSE) between the layer particle size distribution and the normalized block particle size distribution layer by layer; Step 1.5: Retrieve the minimum value in the RMSE sequence obtained in step 1.4 and record its corresponding layer number as the marker layer of the adjacent state of the target pair; Step 1.6: Record the marker layer number obtained in step 1.5 into the metadata area of the flash memory for later use in the data storage cycle; Step 2: During the data storage cycle, the threshold voltage distribution of the block granularity is quickly obtained by sensing the mark layer; In step 2, after executing two rounds of step 1.6, the block granularity predicted by the marker layer is used to optimize the read reference voltage for reading to reduce the number of bit errors; Step 3: Obtain the optimized read reference voltage based on the threshold voltage distribution of the block granularity.

2. The method according to claim 1, wherein: In step one, Assume that the adjacent state pairs to be obtained are X state and Y state respectively. For the layer numbered i, the distributions of the X state and the Y state are f Xi (ν), and f Yi (ν), For the entire block, the distributions of the X state and the Y state are g Xi (v), and g Yi (v), The actual distribution is normalized by the number of layers to obtain the block size distribution.

3. The method according to claim 2, wherein: The root mean square error is used to measure the distribution difference between a layer and a block. The calculation formula is: Where N is the number of sampling points of X and Y states; The marker layer is the layer that satisfies the minimum RMSE. For adjacent X states and Y states, the distribution of the marker layer is closest to the normalized distribution of the entire block.

4. A 3D NAND flash memory block granularity threshold voltage distribution sensing device based on a marking layer, characterized by: The device includes: an offline module, an online module and a voltage management module An offline module is used to obtain the mark layer of each adjacent state pair in the block granularity offline at the initial stage of data storage and record it in the storage metadata area; The offline module also includes: A data writing module, used for writing data and applying residence loss stress to allow the stored data to overcome the early unstable rapid residence loss stage and exhibit a long-term residence loss error mode; The state acquisition module is used to obtain the distribution of the target to the adjacent states layer by layer; Normalization processing module, used to obtain the distribution of the target of the entire block to the adjacent state, and divide it by the number of layers to obtain the normalized block granularity distribution; Root mean square error calculation module, layer by layer calculation of the root mean square error RMSE between the layer particle size distribution and the normalized block particle size distribution; The retrieval module is used to retrieve the minimum value in the RMSE sequence obtained by the root mean square error calculation module, and record the corresponding layer number as the marker layer of the adjacent state of the target pair; The recording module is used to record the marked layer numbers obtained in the retrieval module into the metadata area of the flash memory for later call during the data storage cycle; The online module is used to quickly obtain the threshold voltage distribution at the block granularity by sensing the marked layer during the data storage cycle; use the predicted block granularity of the marked layer to optimize the read reference voltage for reading and reduce the number of error bits; The voltage management module is used to obtain the optimized read reference voltage according to the threshold voltage distribution at the block granularity.

5. The device according to claim 4, wherein: Suppose the adjacent state pairs to be obtained are the X state and the Y state respectively, For the layer numbered i, the distributions of the X state and the Y state are f Xi (ν), and f Yi (ν), For the entire block, the distributions of the X state and the Y state are g Xi (v), and g Yi (v), Then the actual distribution is normalized by dividing by the number of layers to obtain the block granularity distribution.

6. An electronic device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 3.

7. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, it implements the steps of the method according to any one of claims 1 to 3.

Images