List of changes
Introduction
High-level block diagram
The previous figure shows a high-level view of the DDR memory subsystem. The system may be seen as 4 different entities. A Memory Controller (MC), a DFI IP, N PHY (I/O Banks), and M DDR memory.
Description
The Double Data Rate (DDR2) Physical Interface (PHY) IP core is a soft IP that provides connectivity and a standard interface between a DDR2 Memory Controller (MC) and DDR2 memory devices compliant with JESD79-2 specification.
The DFI protocol defines the signals and their relationships to transfer commands and data between a standardized DFI interface and DDR2 devices.
The functionality, configuration and compatibility of the DDR2 PHY IP core are described in this documentation.
Features
The NanoXplore DDR2 PHY IP core provides the following features:
Interfacing with industry standard DDR2 SDRAM devices
DDR2 initialization compliant with JESD79-2 with user programmable parameters
Voltage / Temperature drifts compensation
DFI PHY Update requests
Support for x4, x8 or x16 devices
Programmable burst order (Sequential / Interleave)
On-chip dynamic terminators (FPGA side)
Supports dynamic ODT
Integration with NanoXplore hardware macro blocks
Dynamic training on start-up
Byte-lane/Nibble-lane independence
DFI Clock Disabling
1:4 Frequency ratio between MC and DFI
Theoretical maximum frequency of 400 MHz / 800 Mbit/s
Support for differential DQS Data Strobe and Clock signals
Memory density up to 4Gb
Support for ECC is possible
Supports self-refresh entry/exit as well as power-down entry/exit
Programmable support for SSTL_1V8_I and SSTL_1V8_II for commands/data
Supports CAS Latency between 2 and 6
Supports Additive Latency between 0 and 4
Additional Documentation
Refer to the additional documentation for more information on the DFI / DDR2 Specifications
Document | Description |
DDR PHY Interface, Version 4.0 | Description of signals and their relationships that make up the DFI Interface |
DDR2 SDRAM Specification JESD79-2E | JEDEC Standard defining the SDRAM DDR2 protocol and functionality. |
NanoXplore Library Guide |
|
NanoXplore FPGA NG-Medium User Guide |
|
DFI IP Core
Functional Description
This IP is responsible for communicating to and from the DDR physical layer correctly to implement the DDR protocol.
It implements the DFI specification v4.0 Control / Write Data / Read Data / Status and Update Interfaces in its current version.
The IP is configurable (via a VHDL package file, see the Configuration section) and may be adapted to different system configurations (Signal widths, number of bytelanes, locations of pins, ...).
It only supports a frequency ratio of 4, which means the controller runs in a clock domain of its own, 4 times slower than the memory clock.
The Update interface won’t acknowledge Update requests coming the Memory Controller. It shall only requests Updates when it needs to adjust its internal delays.
The DFI IP realizes a JESD79-2E compliant memory initialization sequence on start-up, and initializes memory Mode Registers with the given parameters. It then calibrates its internal logic by realizing two accesses (one burst of 2xBL8 Read and one burst of 1 BL8 Read).
For more information on DFI timing parameters implemented, refer to section II.F, “Timing Parameters”.
The NanoXplore DDR PHY IP core requires all read commands to be sent on phase 0 out of 3, independently of the chosen CAS Latency or Additive Latency; it also requires using burst lengths of 8.
Implementation
The DFI IP implements an internal DFI IP Core, which uses a 1:2 frequency ratio between its work clock and the DDR clock.
The rest of the IP consists of CDCs, and routing between the actual MC-DFI Interface and the DFI IP Core, necessary to fully implement a frequency ratio of 4.
The physical layers of the DFI IP works using only positive edges; thus, requiring a clock of frequency twice as fast as the frequency of the DDR memory device clock.
Configuration
The DFI IP is configurable via the VHDL package pinout_pack.
Configurations include widths of signals, physical locations of pins, and number of banks used.
The same reasoning applies for the CK / CK# pins and the dfi_dram_clk_disable signal.
Users should configure properly the DFI IP using this package, and take care that all arrays (DQ, AD, BA, …) declared in this VHDL package are complete and coherent.
Bank and pad indexes during pad declaration are using logical locations.
E.G:
constant DQ2 : pad_t := (1, 9);
This line means pad DQ2 is on Bank1 (logical index), and occupies pad #9 (physical) (corresponding to pad D05N).
type array_iob is array (0 to NB_IOB-1) of string; -- IOB
constant IDX_IOB : array_iob := ("IOB10", "IOB11");
The above line makes a correspondence between physical and logical bank indexes.
Here, Bank0 would use physical bank IOB10, and Bank1 would use IOB11; hence pad DQ2 would be located on pad IOB11D05N.
The pinout_pack configuration VHDL package file must contain the following constants:
Name | Type | Description | Affects width of |
NB_IO | Natural | Number of I/Os per Bank, 30 for ng-medium |
|
NB_IOB | Natural | Number of I/O banks the DDR Interface spans over. |
|
|
|
|
|
NB_DQ | Natural | Number of DQ pins | dfi_wrdata_i (twice) dfi_rddata_o (twice) ddr_dq_io |
NB_DQS | Natural | Number of DQSP pins | dfi_wrdata_en_i dfi_rddata_en_i dfi_rddata_valid_o dfi_data_byte_disable_i ddr_dqs_io |
NB_DM | Natural | Number of Data Mask pins | dfi_wrdata_mask_i ddr_dm_o |
NB_AD | Natural | Number of address pins
| dfi_address_i ddr_ad_o |
NB_BA | Natural | Number of Bank Address pins | dfi_bank_i ddr_ba_o |
NB_CKE | Natural | Number of CKE pins | dfi_cke_i ddr_cke_o |
|
|
|
|
NB_ODT | Natural | Number of ODT pins | dfi_odt_i ddr_odt_o |
NB_CS | Natural | Number of CS pins | dfi_cs_n_i ddr_cs_o |
NB_CK | Natural | Number of CKP pins | dfi_dram_clk_disable_i ddr_ck_o |
NB_CAS | Natural | Number of CAS pins | dfi_cas_n_i ddr_cas_o |
NB_RAS | Natural | Number of RAS pins | dfi_ras_n_i ddr_ras_o |
NB_WEN | Natural | Number of WEN pins | dfi_wen_n_i ddr_wen_o |
NB_RST | Natural | Number of Reset pins (DDR3 only) | dfi_reset_n_i ddr_rst_o |
These signal widths shall be reflected at the DFI Interface.
I.E. Using NB_RAS = 2 means all 4 phase-replicated dfi_ras_pN shall have a width of 2.
If the user wishes to duplicate a pin, this duplication shall be reflected at the DFI Interface.
The user should then fill the following arrays with the corresponding pads:
Name | Number of elements | Type of elements | Description |
IDX_IOB | NB_IOB | string | Correspondence between logical indexing and actual banks used in the FPGA |
DQ | NB_DQ | pad_t |
|
DQSP | NB_DQS | pad_t |
|
DQSN | NB_DQS | pad_t |
|
DM | NB_DM | pad_t |
|
AD | NB_AD | pad_t |
|
BA | NB_BA | pad_t |
|
CKP | NB_CK | pad_t |
|
CS | NB_CS | pad_t |
|
ODT | NB_ODT | pad_t |
|
CKE | NB_CKE | pad_t |
|
CAS | NB_CAS | pad_t |
|
RAS | NB_RAS | pad_t |
|
WEN | NB_WEN | pad_t |
|
pad_t is a custom VHDL type declared inside the package;
type pad_t is record
bank : natural; -- Index of bank
pad : natural; -- Index of pad
end record pad_t;
Configuration package file of Evaluation Kit DK625v* shall be included as an example.
Users shall not re-declare the ddr_* HDL pins in their pads.py configuration file.
The following table makes a correspondence between Logical Indexes and their physical locations:
Generics
Name | Type | Default Value | Description |
calClockDivPowerOf2 | Integer | 0 | By which power of 2 to divide ck_ddr for it to fall within 100-200 MHz |
invertOutputCK | boolean | false | Inverts Clock output. Set to true on DK625V*. |
bypassInit | boolean | false | Set to true if using a clock ratio other than 4, or if you wish to let the memory controller realize DDR initialization Refer to section « 4 Limitations and compatibility with DFI 4.0 » if using true |
|
|
|
|
tCK | Integer | 8000 | DDR tCK in picoseconds |
tRPA | Integer | 25000 | DDR tRPA in picoseconds |
tRFC | Integer | 327000 | DDR tRFC in picoseconds |
tRCD | Integer | 20000 | DDR tRCD in picoseconds |
|
|
|
|
casLatency | Integer | 5 | DDR Cas Latency to be set during initialization |
writeRecovery | Integer | 2 | DDR Write Recovery (in cycles) |
sequentialBurst | Boolean | True | Used during mode register initialization |
rttNomnial | Integer | 75 | Shall be 50, 75 or 150. Used during mode register initialization |
differentialDQS | Boolean | True | True to configure and use differential DQS |
outputDriveImpendanceCtrl | Boolean | True | True for full strength False for reduced strength |
additiveLatency | Integer | 0 | Additive Latency |
|
|
|
|
simu | Boolean | false | Cuts down wait times during initialization. |
fpgaDqTerm | String | “75” | Termination value to be used for FPGA DQ pads |
fpgaDqsTerm | String | “75” | Termination value to be used for FPGA DQS pads |
DqsDqDmPadType | String | “SSTL_1.8V_I” | Shall be set to “SSTL_1.8V_I” or “SSTL_1.8V_II” |
CmdPadType | String | “SSTL_1.8V_I” | Shall be set to “SSTL_1.8V_I” or “SSTL_1.8V_II” |
DqsDqDmSlewRate | String | “Fast” | SlewRate on Data pads. Fast is recommended. |
CmdSlewRate | String | “Fast” | SlewRate on Command pads. Fast is recommended. |
DataTurbo | String | “True” | DQ/DQS Input pad config. True is recommended. |
DFI Interface
All signals suffixed by _pN or by _wN are replicated into phase-specific signals (p0, p1, p2, p3)
Refer to DFI Specification 4.0 for more information on these signals.
Control Interface
Name | Direction | Width | Description |
dfi_address_pN_i | in | DFI_ADDRESS_WIDTH |
|
dfi_bank_pN_i | in | DFI_BANK_WIDTH |
|
dfi_cas_n_pN_i | in | NB_CAS |
|
dfi_cke_pN_i | in | NB_CKE |
|
dfi_cs_n_pN_i | in | NB_CS |
|
dfi_odt_pN_i | in | NB_ODT |
|
dfi_ras_pN_i | in | NB_RAS |
|
dfi_we_n_pN_i | in | NB_WEN |
|
dfi_reset_n_pN_i | in | NB_RST-1 | DDR3 Only. |
Write Data Interface
Name | Direction | Width | Description |
dfi_wrdata_pN_i | in | DFI_DATA_WIDTH |
|
dfi_wrdata_en_pN_i | in | DFI_DATA_ENABLE_WIDTH |
|
dfi_wrdata_mask_pN_i | in | DFI_DM_WIDTH | One bit per bytelane |
Read Data Interface
Name | Direction | Width | Description |
dfi_rddata_en_pN_i | in | DFI_DATA_ENABLE_WIDTH |
|
dfi_rddata_wN_o | out | DFI_DATA_WIDTH |
|
dfi_rddata_valid_wN_o | out | DFI_READ_DATA_VALID_WIDTH |
|
Update Interface
Name | Direction | Width | Description |
dfi_ctrlupd_req_i | in | 1 | NC |
dfi_ctrlupd_ack_o | out | 1 | Always ‘0’ |
dfi_phyupd_req_o | out | 1 |
|
dfi_phyupd_type_o | out | 2 | NC – Always “00” |
dfi_phyupd_ack_i | in | 1 |
|
The Update interface in only used when the DFI IP needs to re-calibrate its internal delay lines. It does so by asserting its dfi_phyupd_req_o output signal, then waits on its dfi_phyupd_ack_i input signal.
Once asserted, it shall internally go through re-calibration. It shall de-assert its dfi_phy_req_o output signal once re-calibration is complete.
Status Interface
Name | Direction | Width | Description |
dfi_data_byte_disable_i | in | DFI_DATA_BYTE_DISABLE_WIDTH |
|
dfi_dram_clk_disable_i | in | NB_CK |
|
dfi_freq_ratio_i | in | 2 | 0b00 = 1:1 0b01 = 1:2 0b10 = 1:4 Only 1:4 frequency ratio is supported |
dfi_init_complete_o | out | 1 |
|
dfi_init_start_i | in | 1 | DFI won’t start unless this signal is asserted. Necessary to specify starting freq. Ratio, and byte lanes disabled |
Additional Inputs
Name | Direction | Width | Description |
ck_sdr_i | in | 1 | SDR clock (F = 2xFDDR) |
ck_dfi_i | in | 1 | DFI Memory Controller clock (F = ¼ of FDDR) |
ck_ddr_i | in | 1 | DDR clock (F = FDDR, or divided by a power of 2 to fall within 100-200 MHz). This clock is only used for delay line calibration. |
ck_work_i | in | 1 | Work clock (F = ½ of FDDR) |
err_o | out | 1 | Error out signal, asserted while re-calibrating, during a DFI |
rst_n_work_i | in | 1 | Reset signal in ck_work clock domain |
rst_n_dfi_i | in | 1 | Reset signal in memory controller clock domain |
Memory Interface
Name | Direction | Width | Description |
ser_ck_o | Out | NB_CK |
|
ser_ad_o | Out | NB_AD |
|
ser_ba_o | Out | NB_BA |
|
ser_cke_o | Out | NB_CKE |
|
ser_odt_o | Out | NB_ODT |
|
ser_cs_o | Out | NB_CS |
|
ser_dqs_io | InOut | NB_DQS |
|
ser_dq_io | InOut | NB_DQ |
|
ser_msk_o | Out | NB_DM |
|
ser_cas_o | Out | NB_CAS |
|
ser_ras_o | Out | NB_RAS |
|
ser_wen_o | Out | NB_WEN |
|
ser_rst_o | Out | NB_RST | Unused in DDR2. |
Command pads use the CmdPadType generic electrical standard, while data (DQS / DQ and DM) pads use the DqsDqDmPadType generic electrical interface.
DQ pads use the fpgaDqTerm termination value, while DQS pads use the fpgaDqsTerm termination value, both of which need to be filled in as strings.
These pads shall not be declared in the project’s pads.py interface file.
Timing parameters
Control Interface
Parameter | Value | Unit | Description |
tctrl_delay | 10 | DFI clock cycles | Specifies the number of DFI clock cycles after an assertion or de-assertion of the DFI control signals that the control signals at the PHY-DRAM interface reflect the assertion or de-assertion. |
Write Data Interface
Parameter | Value | Unit | Description |
tphy_wrdata | 3 | DFI PHY clock cycles | Specifies the number of DFI PHY clock cycles between when the dfi_wrdata_en signal is asserted to when the associated write data is driven on the dfi_wrdata signal |
tphy_wrdelay | 0 | DFI PHY clock cycles | Specifies the number of DFI PHY clock cycles of additional delay that the PHY must insert between the write data enable and write data once data has been captured from the DFI bus. |
tphy_wrlat | AL+CL-4 = WL-3 | DFI PHY clock cycles | Specifies the number of DFI PHY clock cycles between when a write command is sent on the DFI control interface and when the dfi_wrdata_en signal is asserted. |
Read Data Interface
Parameter | Value | Unit | Description |
tphy_rdlat | 35 | DFI PHY clock cycles | Specifies the maximum number of DFI PHY clock cycles allowed from the assertion of the dfi_rddata_en signal to the assertion of the dfi_rddata_valid signal |
trddata_en | AL+CL-2 = RL-2 | DFI PHY clock cycles | Specifies the number of DFI PHY clock cycles from the assertion of a read command on the DFI to the assertion of the dfi_rddata_en signal. |
Status Interface
Parameter | Value – max | Unit | Description |
tdram_clk_disable | 10 | DFI clock cycles | Specifies the number of DFI clock cycles from the assertion of the dfi_dram_clk_disable signal on the DFI until the clock to the DRAM memory devices, at the PHY-DRAM boundary, maintains a low value. If the DFI clock and the memory clock are not phase-aligned, this timing parameter should be rounded up to the next integer value. |
tdram_clk_enable | 10 | DFI clock cycles | Specifies the number of DFI clock cycles from the de-assertion of the dfi_dram_clk_disable signal on the DFI until the first valid rising edge of the clock to the DRAM memory devices, at the PHY-DRAM boundary. If the DFI clock and the memory clock are not phase-aligned, this timing parameter should be rounded up to the next integer value |
FPGA resource utilization
The following chart gives an approximation of the number of resources used by the NG-Medium DFI IP depending on the memory interface width (e.g. number of DQ pins).
These configurations were tested on NanoXmap 2.9.4, using useCDCForMC = false and bypassInit = false. Different results may be expected if using different parameters.
Limitations and compatibility with DFI 4.0 Spec.
DFI Frequency ratio change protocol is not supported.
Only a DFI Frequency Ratio of 4 is supported
Read commands shall only be sent on phase p0.
The user should take care of using NanoXmap’s PrePlace feature on the following registers:
◦ <your_hierarchy>|<dfi_instance_name>|inst_top_dfi|ins_dfi|ins_dfi_core|dfi_rddata_w0_o_reg[*]
◦ <your_hierarchy>|<dfi_instance_name>|inst_top_dfi|ins_dfi|ins_dfi_core|dfi_rddata_w1_o_reg[*]
◦ <your_hierarchy>|<dfi_instance_name>|inst_top_dfi|ins_dfi|ins_dfi_core|dqs_r_reg[0-3]
These registers should be placed as closely as possible to the corresponding IO Banks.
Even if the user wishes to not use differential DQS, the user shall not use the DQSN pad for a different signal.
When setting the generic bypassInit to true, extra care should be taken after realizing DDR initialization.
Once the initialization phase is over; the user should realize 3 BL8 Read accesses.
The first two shall be ‘back-to-back’. The last one should be on the same phase (p0, ..., p3) on the first ones.
All following read commands shall be sent on the same phase (p0, ..., p3) as the first ones.
There is no guarantee that these read accesses will return the expected data, and they won’t generate any rddata_valid signals.
Product Support
NanoXplore backs its products with support services, including Customer Technical Support.
Customer Technical Support
If you encounter any problems with the DDR2 PHY IP Core or with any NanoXplore products, please contact our team of highly skilled engineers. You may contact them by emailing support@nanoxplore.com
Make sure to include your credentials, including your full name and company name when contacting our support team.