在FPGA中，大规模数据的存储常常会用到DDR。为了方便用户使用，Xilinx提供了DDR MIG IP核，用户能够通过AXI接口进行DDR的读写访问，然而MIG内部自动实现了许多环节，不利于用户深入理解DDR的底层逻辑。
本文以美光(Micron）公司生产的DDR3芯片MT41J512M8RH-093为例，说明DDR芯片的操作过程。
DDR芯片的使用关键在于令接口的信号变化满足时序要求，在初始化过程中主要关注下面几个时序参数（来自P33 Table 9: Timing Parameters Used for IDD Measurements -- Clock Units与P96 Table 59: Electrical Characteristics and AC Operating Conditions for Speed Extensions (Continued)）
* CK(MIN) 0.938 ns
* CL 14 CK
* RCD(MIN) 14 CK
* RC(MIN) 50 CK
* RAS(MIN) 36 CK
* RP(MIN) 14 CK
* FAW 27 CK
* RRD 6 CK
* RFC 279CK
* XPR > max(5CK, RFC+10ns)
* MRD > 4 CK
* MOD > max(12 CK, 15ns)
* ZQinit < max(512nCK, 640ns)
* DLLK > 512 CK
从datasheet的P12页的Fig2. Simplified State Diagram可以看到，DDR3芯片在上电（Power applied）后需要经过一系列的初始化步骤（主要包含三个部分Reset Procedure、Initialization、ZQ Calibration），之后进入正常工作状态（idle）。
上图中Command由多个信号的变化构成，在初始化过程主要用到以下几个指令。（来自P118 Table 70: Truth Table -- Command）
* COMMAND | NOP | MRS_1 | MRS_2 | MRS_3 | MRS_4 | ZQCL
* ddr_cke | 1 1 | 1 1 | 1 1 | 1 1 | 1 1 | 1 1
* ddr_dqs_en | 0 | 0 | 0 | 0 | 0 | 0
* ddr_dq_en | 0 | 0 | 0 | 0 | 0 | 0
* ddr_cs_n | 0 | 0 | 0 | 0 | 0 | 0
* ddr_ras_n | 1 | 0 | 0 | 0 | 0 | 1
* ddr_cas_n | 1 | 0 | 0 | 0 | 0 | 1
* ddr_we_n | 1 | 0 | 0 | 0 | 0 | 0
* ddr_ba | vvv | 010 | 011 | 001 | 000 | 000
* ddr_addr | vvv | 28 | 00 | 44 | 124 | a[10]=1
* ddr_odt | 0 | 0 | 0 | 0 | 0 | 0
这里将时钟周期取为最小时钟周期0.938ns，对应时钟频率1066.099MHz。经过计算，Initialization阶段每条指令执行后的等待时间均不超过1us，因此这里将Initialization阶段每条指令执行后的等待时间均简化1us，最终得到的DDR3初始化代码如下：
c
`timescale 1ns / 1ps
//
// Company:
// Engineer: wjh776a68
//
// Create Date: 01/05/2024 09:45:15 AM
// Design Name: micron_ddr
// Module Name: micron_ddr_init
// Project Name: micron_ddr
// Target Devices: vu9p
// Tool Versions: 2017.4
// Description:
//
// Dependencies:
//
// Revision:
// Revision 0.01 - File Created
// Additional Comments:
//
//
// ddr3 x8 4Gb MT41J512M8RH-093
/****************************
* DDR3L-2133 https://www.micron.com/-/media/client/global/documents/products/data-sheet/dram/ddr3/4gb_ddr3l.pdf?rev=305217e2f9bd4ef48d7c6f353dfc064c
* CK(MIN) 0.938 ns
* CL 14 CK
* RCD(MIN) 14 CK
* RC(MIN) 50 CK
* RAS(MIN) 36 CK
* RP(MIN) 14 CK
* FAW 27 CK
* RRD 6 CK
* RFC 279CK
* XPR >max(5CK, RFC+10ns)
* MRD >4CK
* MOD >max(12CK, 15ns)
* ZQinit <max(512nCK, 640ns)
* DLLK >512CK
* command all p118
* initial waveform p137
*
* COMMAND | NOP | MRS_1 | MRS_2 | MRS_3 | MRS_4 | ZQCL
* ddr_cke | 1 1 | 1 1 | 1 1 | 1 1 | 1 1 | 1 1
* ddr_dqs_en | 0 | | | | |
* ddr_dq_en | 0 | | | | |
* ddr_cs_n | 0 | 0 | 0 | 0 | 0 | 0
* ddr_ras_n | 1 | 0 | 0 | 0 | 0 | 1
* ddr_cas_n | 1 | 0 | 0 | 0 | 0 | 1
* ddr_we_n | 1 | 0 | 0 | 0 | 0 | 0
* ddr_ba | vvv | 010 | 011 | 001 | 000 |
* ddr_addr | vvv | 28 | 00 | 44 | 124 | a[10]
* ddr_odt | 0 | | | | |
***************************************************************/
module micron_ddr_init #(
parameter CLK_FREQ = 1066.099, // MHz
parameter _1MS_CYCLE = 10.0**-3 / (1.0 / (CLK_FREQ * 10**6)),
parameter _1US_CYCLE = 10.0**-6 / (1.0 / (CLK_FREQ * 10**6)),
parameter integer INITIAL_CYCLE = 200 * _1US_CYCLE,
parameter integer INITIAL_STABLE_CYCLE = 500 * _1US_CYCLE,
parameter integer FREE_CYCLE = _1US_CYCLE
) (
output reg [15:0] ddr_addr,
output reg [2:0] ddr_ba,
output reg ddr_cas_n,
output reg [0:0] ddr_ck_n,
output reg [0:0] ddr_ck_p,
output reg [0:0] ddr_cke,
output reg [0:0] ddr_cs_n,
output reg [0:0] ddr_dm,
inout [7:0] ddr_dq,
inout [0:0] ddr_dqs_n,
inout [0:0] ddr_dqs_p,
output reg [0:0] ddr_odt,
output reg ddr_ras_n,
output reg ddr_reset_n,
output reg ddr_we_n,
input clk
);
localparam [27:0] NOP_CMD = {11'b11000111000, 16'h0000, 1'b0};
localparam [27:0] MRS1_CMD = {11'b11000000010, 16'h0028, 1'b0};
localparam [27:0] MRS2_CMD = {11'b11000000011, 16'h0000, 1'b0};
localparam [27:0] MRS3_CMD = {11'b11000000001, 16'h0044, 1'b0};
localparam [27:0] MRS4_CMD = {11'b11000000000, 16'h0124, 1'b0};
localparam [27:0] ZQCL_CMD = {11'b11000110000, 16'h0400, 1'b0};
reg ddr_cke_p1, ddr_cke_p2;
reg ddr_dqs_i, ddr_dqs_o, ddr_dqs_en;
reg ddr_dq_i, ddr_dq_o, ddr_dq_en;
OBUFDS OBUFDS_ck (
.O(ddr_ck_p), // 1-bit output: Diff_p output (connect directly to top-level port)
.OB(ddr_ck_n), // 1-bit output: Diff_n output (connect directly to top-level port)
.I(clk) // 1-bit input: Buffer input
);
IOBUFDS #(
.DQS_BIAS("FALSE") // (FALSE, TRUE)
)
IOBUFDS_dqs_inst (
.O(ddr_dqs_o), // 1-bit output: Buffer output
.I(ddr_dqs_i), // 1-bit input: Buffer input
.IO(ddr_dqs_p), // 1-bit inout: Diff_p inout (connect directly to top-level port)
.IOB(ddr_dqs_n), // 1-bit inout: Diff_n inout (connect directly to top-level port)
.T(ddr_dqs_en) // 1-bit input: 3-state enable input
);
IOBUF IOBUF_dq_inst (
.O(ddr_dq_o), // 1-bit output: Buffer output
.I(ddr_dq_i), // 1-bit input: Buffer input
.IO(ddr_dq), // 1-bit inout: Buffer inout (connect directly to top-level port)
.T(ddr_dq_en) // 1-bit input: 3-state enable input
);
ODDRE1 #(
.IS_C_INVERTED(1'b1), // Optional inversion for C
.IS_D1_INVERTED(1'b0), // Unsupported, do not use
.IS_D2_INVERTED(1'b0), // Unsupported, do not use
.SRVAL(1'b0) // Initializes the ODDRE1 Flip-Flops to the specified value (1'b0, 1'b1)
)
ODDRE1_cke_inst (
.Q(ddr_cke), // 1-bit output: Data output to IOB
.C(clk), // 1-bit input: High-speed clock input
.D1(ddr_cke_p1), // 1-bit input: Parallel data input 1
.D2(ddr_cke_p2), // 1-bit input: Parallel data input 2
.SR(1'b0) // 1-bit input: Active High Async Reset
);
//OBUFDS OBUFDS_dqs (
// .O(ddr_dqs_p), // 1-bit output: Diff_p output (connect directly to top-level port)
// .OB(ddr_dqs_n), // 1-bit output: Diff_n output (connect directly to top-level port)
// .I(ddr_dqs) // 1-bit input: Buffer input
// );
reg [15:0] ddr_addr_r;
reg [2:0] ddr_ba_r;
reg ddr_cas_n_r;
reg [0:0] ddr_cs_n_r;
reg [0:0] ddr_dm_r; // no ref
reg [0:0] ddr_odt_r;
reg ddr_ras_n_r;
reg ddr_we_n_r;
reg ddr_dq_en_r;
reg ddr_dqs_en_r;
initial begin
{ddr_cke_p2, ddr_cke_p1, ddr_dqs_en, ddr_dq_en, ddr_cs_n, ddr_ras_n, ddr_cas_n, ddr_we_n, ddr_ba, ddr_addr, ddr_odt} <= NOP_CMD;
{ddr_cke_p2, ddr_cke_p1, ddr_dqs_en_r, ddr_dq_en_r, ddr_cs_n_r, ddr_ras_n_r, ddr_cas_n_r, ddr_we_n_r, ddr_ba_r, ddr_addr_r, ddr_odt_r} <= NOP_CMD;
end
always @(negedge clk) begin
ddr_addr <= ddr_addr_r ;
ddr_ba <= ddr_ba_r ;
ddr_cas_n <= ddr_cas_n_r ;
ddr_cs_n <= ddr_cs_n_r ;
ddr_dm <= ddr_dm_r ;
ddr_odt <= ddr_odt_r ;
ddr_ras_n <= ddr_ras_n_r ;
ddr_we_n <= ddr_we_n_r ;
ddr_dq_en <= ddr_dq_en_r;
ddr_dqs_en <= ddr_dqs_en_r;
end
reg [5:0] cs = 0, ns;
reg [31:0] initial_cnt = 0;
reg [31:0] initial_stable_cnt = 0;
reg [31:0] freerun_cnt = 0;
reg [3:0] initial_cmd_ptr = 0;
reg [27:0] initial_cmd_seq[0:5] = '{NOP_CMD, MRS1_CMD, MRS2_CMD, MRS3_CMD, MRS4_CMD, ZQCL_CMD};
reg initial_finish = 0;
always @(negedge clk) begin
cs <= ns;
end
always @(*) begin
case (cs)
0: begin
if (initial_cnt == INITIAL_CYCLE) begin // wait 200us
ns = 1;
end else begin
ns = 0;
end
end
1: begin // wait 500us
if (initial_stable_cnt == INITIAL_STABLE_CYCLE) begin // wait 500us
ns = 2;
end else begin
ns = 1;
end
end
2: begin
ns = 3;
end
3: begin
if (freerun_cnt == FREE_CYCLE) begin
if (initial_finish) begin
ns = 4;
end else begin
ns = 2;
end
end else begin
ns = 3;
end
end
4: begin
// finish initial, enter idle state
end
default: begin
ns = 0;
end
endcase
end
always @(negedge clk) begin
case (ns)
0: begin
initial_cnt <= initial_cnt + 1;
end
default: begin
initial_cnt <= 0;
end
endcase
end
always @(negedge clk) begin
case (ns)
1: begin
initial_stable_cnt <= initial_stable_cnt + 1;
end
default: begin
initial_stable_cnt <= 0;
end
endcase
end
always @(negedge clk) begin
case (ns)
3: begin
freerun_cnt <= freerun_cnt + 1;
end
default: begin
freerun_cnt <= 0;
end
endcase
end
always @(negedge clk) begin
case (ns)
2: begin
if (initial_cmd_ptr == 6 - 1) begin
initial_finish <= 1;
end else begin
initial_finish <= 0;
end
initial_cmd_ptr <= initial_cmd_ptr + 1;
end
endcase
end
always @(negedge clk) begin
case (ns)
0: begin
ddr_reset_n <= 1'b0;
{ddr_cke_p2, ddr_cke_p1} <= 2'b0;
ddr_dqs_en_r <= 1'b0;
ddr_dq_en_r <= 1'b0;
end
1: begin
ddr_reset_n <= 1'b1;
{ddr_cke_p2, ddr_cke_p1} <= 2'b0;
ddr_dqs_en_r <= 1'b0;
ddr_dq_en_r <= 1'b0;
end
2: begin
{ddr_cke_p2, ddr_cke_p1, ddr_dqs_en_r, ddr_dq_en_r, ddr_cs_n_r, ddr_ras_n_r, ddr_cas_n_r, ddr_we_n_r, ddr_ba_r, ddr_addr_r, ddr_odt_r} <= initial_cmd_seq[initial_cmd_ptr];
end
3: begin
{ddr_cke_p2, ddr_cke_p1, ddr_dqs_en_r, ddr_dq_en_r, ddr_cs_n_r, ddr_ras_n_r, ddr_cas_n_r, ddr_we_n_r, ddr_ba_r, ddr_addr_r, ddr_odt_r} <= NOP_CMD;
end
default: begin
ddr_reset_n <= 1'b1;
end
endcase
end
endmodule上述代码已在Vivado 2017.4中进行了仿真测试，可替换ddr示例工程中的example_top自行仿真。
