Posts

FPGAs are optimised for massive parallelism, but their on-chip memory resources are limited. Block RAM (BRAM) and UltraRAM (URAM) are fast and easy to use, but they typically provide only a few tens of megabits to a few hundred megabits of capacity. When your application needs gigabytes — high-resolution video frames, machine-learning weight tables, large look-up tables — external DDR memory becomes essential. Why External DDR Memory? Memory Type Location Typical Capacity Bandwidth Access Latency Distributed RAM On-chip (LUT-based) < 1 MB Very high 1 cycle Block RAM (BRAM) On-chip (dedicated) up to ~200 MB High 1–2 cycles UltraRAM (URAM) On-chip (UltraScale+) up to ~500 MB High 2–3 cycles External DDR Off-chip (dedicated chip) 1–16+ GB Medium-High ~50–100 ns For applications such as: ...

What Happened to FPGA Prices? Between 2020 and 2022, the semiconductor industry experienced what analysts called “chipflation” — a sustained, broad-based price increase driven by pandemic-related demand spikes, supply chain disruptions, and surging investment in AI and 5G infrastructure. FPGA manufacturers were not immune. AMD (Xilinx) and Intel (Altera) both cited rising TSMC wafer costs as justification for significant list-price increases — in some cases close to double the pre-pandemic price. Unlike commodity memory or microcontroller price spikes that eventually normalised, elevated FPGA prices have proven sticky. ...

In digital communication, data is represented as a binary bitstream of 0s and 1s. In practice, real-world data is rarely truly random — it often contains repetitive patterns, especially long runs of identical bits (e.g., ...000000... or ...111111...). At low speeds this causes no real harm, but at 1 Gbps and beyond, such patterns create critical signal integrity problems. At 1 Gbps, one bit period is just 1 nanosecond, so a run of tens or hundreds of identical bits means the signal voltage stays constant for a relatively long time. ...

디지털 데이터는 본질적으로 이진 비트(0과 1)의 집합이지만, 이를 실제 물리적 전송 매체(구리선, 광섬유 등)를 통해 송수신하려면 전기적 또는 광학적 신호로 변환하는 과정이 필수적이다. 이 변환 과정은 단순히 비트를 전압 레벨로 바꾸는 것 이상의 복잡한 기술적 과제를 수반한다. 가장 주요한 두 가지 과제는 DC 성분(직류 성분)의 축적과 수신기 클럭 동기화이다. 동일한 비트(예: ‘00000…’ 또는 ‘11111…’)가 장시간 연속되면 전압이 한쪽 극성으로만 유지되어 신호에 직류 성분이 누적된다. 또한, 신호의 전이(transition)가 사라져 수신기 측의 위상 동기 루프(Phase-Locked Loop, PLL) 회로가 클럭을 추출하지 못해 데이터 비트의 경계를 식별하는 데 실패한다. 이러한 문제들은 데이터 전송의 신뢰성과 무결성을 심각하게 저해한다. ...

The Limits of Parallel Buses Traditional parallel data buses were efficient at low clock rates, but as clock frequencies increased they ran into fundamental physical barriers. The core problem is timing skew: with many parallel data lanes plus a separate clock lane, each signal travels a slightly different path length on the PCB and through the package, arriving at the receiver at slightly different times. As the clock period shrinks, even a small skew becomes a significant fraction of a bit period, and data errors follow. ...

UART Frame Format UART (Universal Asynchronous Receiver/Transmitter) sends data in a structured frame that includes the following components: Start Bit: Always 0; indicates the beginning of the frame. Data Bits: The actual payload, typically 5 to 8 bits. Parity Bit (optional): Used for simple error detection. Stop Bit(s): One or two bits set to 1; mark the end of the frame. Parity Bit Modes Even Parity: The parity bit is set such that the total number of 1s (including the parity bit) is even. ...

https://docs.amd.com/r/2021.1-English/ug949-vivado-design-methodology/When-and-Where-to-Use-a-Reset Advantages of Synchronous Resets BenefitExplanationBetter Resource MappingSynchronous resets can be mapped directly to more FPGA resources (e.g., flip-flops inside DSPs, BRAMs).Improved Logic PerformanceAsynchronous resets can negatively affect general logic timing paths and increase routing complexity.Simpler RoutingA global asynchronous reset might not increase control sets but does require routing reset wires to many elements, which can become complex.Protects Memory StructuresAsynchronous resets may corrupt the contents of BRAMs, LUTRAMs, and SRLs during reset assertion—especially if those resources are driven by asynchronously reset registers.Placement FlexibilityWith synchronous resets, the logic can be more easily remapped during implementation for better packing and performance.Compatibility with Dedicated BlocksSome blocks like DSP48s and BRAMs only support synchronous resets, so using asynchronous resets can prevent proper inference into those blocks. ...

In HDL design (such as Verilog or VHDL), an active-low reset means the reset signal is enabled when it is low (logic 0). It is widely used in both ASIC and FPGA designs for several practical and electrical reasons: 1. Electrical Stability Logic low (0 volts, GND) is generally more stable and noise-immune than logic high (VCC). During power-up, signals tend to default to low due to internal pull-downs or the absence of a valid voltage, so using an active-low reset ensures that the system starts in a known, reset state. ...

What is the RoCE v2? RoCEv2, or RDMA over Converged Ethernet version2 is a network protocol that enable Remote Direct Memory Access (RDMA) over standard Ethernet networks using UDP/IP. It’s designed to deliver high-throughput, low-latency communication while minimizing CPU usage-making it ideal for data centers, high-performance computing, and storage systems. The link below is a lecture on RDMA. It provides a lot of useful information. https://www.coursera.org/learn/the-fundamentals-of-rdma-programming/home/info RoCEv2 Packet Format Ethernet Header Standard Ethernet frame header. ...

What is Skew? Skew refers to the timing difference in signal arrival between multiple paths that are supposed to be synchronized. shows a skewed waveform between data lines. Why Skew Simulation Is Necessary If you simulate your design assuming everything is perfectly aligned, you may encounter issues during actual testing. In real PCBs, skew naturally exists, which is why techniques like length matching are necessary. However, even with length matching, the impact of skew becomes more significant as signal speed increases. Therefore, it’s important to address skew issues during simulation. ...