Memory Array Architectures

A SRAM Cell consists basically of 2 cross coupled CMOS inverters. By applying a logical 1 to the Wordline WL the internal values become visible at the Bitlines BL. The Bitline BL will have the value of the cell while the Bitline \overline{BL} will have the complementary value. The reason is that if there is noise, the noise will influence both lines but will not change the voltage difference. Thats not the only reason, when writing a SRAM Cell we just can write a 0 and dependent on which side we do we set the value, this is due to the Vt drop at the wordline transistores. There are a lot more things to consider when dealing with SRAM Cells but as we are discussing the array architecture here I will skip this.

SRAM-Cell

SRAM-Cell

A sense amplifier will have both bitlines as input and will return a single bit which is representing the value stored in the cell. Memory arrays are organized as row and columns (like a table). Rows are accessed using the Wordline, Columns are accessed by selecting a Bitline. In order to select a row or column we need address encoders.

SRAM Architecture

As you can see we have 3 bits for column addressing and 3 bits for row addressing. With 6 bits we can address 2^6=64 SRAM-Cells (Bits). As you can see the SRAM architecture has a big transistor count, therefore it requires a lot of chip area and it also draws more power than eg. DRAM. But it is very fast and therefore used as Cache on the CPU of PCs. Also microcontrollers use SRAM as they usually do not have external RAM. The user is not interested in the internal architecture of the memory and also the user does not want to care about rows and columns. We simply define that the first 3 bits of the address define the row whereas the last 3 bits define the column. If we do so we end up with a table like this:

111000 111001 111010 111011 111100 111101 111110 111111
110000 110001 110010 110011 110100 110101 110110 110111
101000 101001 101010 101011 101100 101101 101110 101111
100000 100001 100010 100011 100100 100101 100110 100111
011000 011001 011010 011011 011100 011101 011110 011111
010000 010001 010010 010011 010100 010101 010110 010111
001000 001001 001010 001011 001100 001101 001110 001111
000000 000001 000010 000011 000100 000101 000110 000111

Read the page about multiplexers to learn more about how they work.

Memory Architectures with multi bit elements

We are usually not interested in addressing single bits, dependent on the surrounding architecture the size of a single memory elements is usually 8/32/64 bit. We introduce a new symbol for a 8bit Memory element:

8bit Memory Element

8bit Memory Element

There are different ways on how to implement the Address Decoder. I want to discuss 2 possibilities here, the combinatorial and the 1 hot (MUX) approach.

Combinatorial Address Decoder

We can simply check if the current address on the bus is ours using a big end gate:

8bit AND

8bit AND

The AND Gate above would activate the memory element if the address is 01100101.

To keep the schematic simple I illustrated a memory array with 4 address bits and therefore 16 memory elements. Every memory element gets the complete address and need to decide if it is addressed or not. For big arrays this causes a big overhead and also comes with the drawback that every memory element needs another address decoder

combinatorial addressing unidirectional datalines memory array

combinatorial addressing unidirectional datalines memory array

1 hot addressing

To overcome the issues with the combinatorial address encoding we can implement the 1 hot addressing which you already have seen with the SRAM Architecture. There are 2 address busses, one for the column and one for the row. Each of those busses has the same width as the original address and just one bit of each bus is allowed to be high at a time which means that we need to additional logic in order to generate those values but do not need complex address encoding inside the memory element as a simple AND gate is enough and all memory elements are now identical.

 

1hot addressing unidirectional datalines memory array

1hot addressing unidirectional datalines memory array

 

 

Bidirectional data lines

In the example above we had dedicated lines for input and output. Those lines can be combined to a bidirectional data bus. We need to make sure that we can switch the memory elements from read to write operation, therefore we need to introduce a switchable non inverting transmission gate:

transmission gate

It is High Z (non conducting) if CTR is low and conducting if Z is high. However, we use this high level symbol for demonstration purposes, on transistor level things like this are unfortunately not that easy.

bidirectional FF

 

Additional we need to tell each memory element if it is in read or write state which requires additional wires.

combinatorial addressing bidirectional datalines memory array

combinatorial addressing bidirectional datalines memory array

 

 

Leave a Reply