Download presentation
Presentation is loading. Please wait.
Published byPauline Hodge Modified over 9 years ago
2
Memory II Computer Architecture and Design Lecture 4
3
Youpyo Hong @ DGU Target System Requirement There are many types of memories. The question “What is the best memory?” is invalid. The right question is “What is the best memory for my application?”. That means you have to understand the requirement of your target 2
4
Youpyo Hong @ DGU IC vs. Board ICs (Integrated Circuit, Chip) can and must run at very high speed with more complex functions, typically. ICs design/fabrication cost is much higher, typically. Now let’s think about what kind of memory you will prefer 3
5
Youpyo Hong @ DGU Strengths and Weakness of Memories SRAMs Async. SRAM : Read/Write operation finishes in access time. (Typ. access time is ~6ns) Sync. SRAM : Read/Write operation synchronized to clock signal. DRAMs SDR SDRAM : Much higher density (meaning lower cost) than Sync. SRAM. High thruput due to burst transaction support. DDR SDRAM : Thruput is even higher than SDR SDRAM because DDR supports burst transaction and double edge sampling. The chips using the external DDR should run twice faster than using external SDR to utilize the performance of DDR. Needs differential clocks. DDR cost is getting cheaper than SDR. 4
6
Youpyo Hong @ DGU Burst Transaction Burst transaction means accessing multiple data in continuous address using pipelining. To setup start address and size for memrory, burst transactions need setup overhead of a few clock cycles. Programs and data (specially for multimedia application) are typically accessed by burst transactions these days. Why? 5
7
Youpyo Hong @ DGU Flash Memory Nor flash is for simple applications. No burst transaction. Thruput lower compared to NAND flash. No access overhead due to pipelining for burst transactions as NAND does. Nand flash is for applications where massive storage or/and high thruput is required. Flash memory is typically implemented off-chip because on-chip flash is hard to implement and your don’t have to include on-chip, typically. 6
8
Youpyo Hong @ DGU Synch. SRAM Verilog Code from a Book always @ (posedege clk) if (!wr) sram[a] <= d_in; assign q = (!rd) ? sram[a] : 4’bz; 7
9
Youpyo Hong @ DGU Commercial Sync. SRAM Verilog Model module sram_16x8 (addr, clk, din, dout, we ); parameter addr_width = 4, word_depth = 16, word_width = 8; input [addr_width-1:0] addr; input [word_width-1:0] din; output [word_width-1:0] dout; input clk, we; reg [word_width-1:0] mem [0:word_depth-1]; reg [word_width-1:0] dout; always @(posedge clk) begin if(!we) mem[addr] <= din[word_width-1:0]; end always @(posedge clk) begin #1 dout <= mem[addr]; end endmodule 8 1
10
Youpyo Hong @ DGU Homework #2 1. Find the fastest commercial Async. SRAM and Sync. SRAM from websites. Compare their access times and latencies. 2. Simulate to write the data 0000 0000 0001 0000 0010. 0000 1111 into the synchronous SRAM from address 0 to address 15, and read the content of the synchronous SRAM from address 0 to address 15. Submit 1) DUT 2) Testbench 3) Simulation Output Waveform. 9
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.