AXI-Stream Skid Buffer

Nvidia

Problem Description

Design an AXI-Stream skid buffer that acts as a one-entry elastic pipeline register between an upstream producer and a downstream consumer. The module accepts data from the upstream side (s_valid, s_data) and forwards it downstream (m_valid, m_data), asserting s_ready to the producer whenever it has room to accept a new beat. The critical constraint is that once the upstream handshake completes (both s_valid and s_ready high), the data must never be lost — even if the downstream consumer is not ready (m_ready = 0) the very next cycle. On reset, all outputs deassert and any buffered data is discarded.

The core problem: Without a skid buffer, a naive pipeline register makes s_ready = m_ready — this creates a combinational path from consumer to producer and, worse, it can drop data when m_ready goes low the cycle after a handshake. The skid buffer solves this by providing a one-entry "skid slot" to absorb exactly one beat of backpressure.

Screenshot 2026-04-26 at 5.40.51 PM
Screenshot 2026-04-26 at 5.43.53 PM

Verilog

Code Editor Access

Log in to write, run and submit your solution to this interview question. It's free.

Output

Output goes here...