Intel C2518 FH8065501516710 Data Sheet

Intel

®

 Atom™ Processor C2000 Product Family for Microserver

Datasheet, Vol. 2 of 3

September 2014

60

Order Number: 330061-002US

The write and read data to and from the integrated memory controller (the internal unit 

in SSA handles I/O requests from the core and other bus agents, like SATA) are 

protected by even parity on each byte lane. Parity per byte lane is used since the 

internal data buffers contain byte write enables to support partial writes from the 

requesting agents.

The DDR3 interface is protected by an ECC code for Single-Bit Error Correction (SEC) 

and Double-Bit Error Detection (DED). An 8-bit ECC code word is stored with every 

8 bytes of data, which can protect a 128-bit wide interface. Since the data width is only 

72 bits, 56 single-bit error syndrome codes are available. 

An ECC DIMM only provides additional storage for redundant information, the actual 

error detection/correction takes place within the SoC memory controller.

One of these single-bit error syndrome codes converts from byte parity to the 8-byte 

ECC. When the write data ECC is generated, the parity is checked on all 8-byte lanes 

and the parity error signal generates the ECC. When checking ECC on a read, the check 

is performed assuming that the SoC D-Unit has no write data parity errors. Any 

uncorrectable error syndrome that is detected on a read results in the generation of 

bad parity for all 8-byte lanes.

When a word is written into ECC protected memory, the ECC bits are computed by a set 

of exclusive OR trees. When the word is read back, the exclusive OR trees use the data 

read from the memory to recompute the ECC. The recomputed ECC is compared to the 

ECC bits read from the memory. Any discrepancy indicates an error. By looking at which 

ECC bits do not match, identify which data or ECC bit is in error, or whether a double-

bit error occurred. The result of this ECC calculation is called the syndrome. If the 

syndrome is zero, no error occurred. If the syndrome is non-zero, the syndrome is used 

 to determine which bits are in error, or if the error is uncorrectable.