Etched leveraged Synopsys ZeBu® emulation technology to validate its purpose-built Sohu AI inference processor by running full-scale Llama 8B workloads in a pre-silicon environment.
By enabling system-level validation across hardware, firmware, memory, and software stacks before tape-out, Etched accelerated software development, improved validation throughput, and reduced deployment risk while increasing confidence in silicon readiness.
What You Will Learn
- How Etched used Synopsys ZeBu Server 5 to execute full LLM inference workloads in emulation and validate complete AI system behavior before silicon availability.
- Best practices for creating a shared pre-silicon development platform that enables hardware, firmware, software, and validation teams to collaborate and accelerate development cycles.
- How full-system emulation can uncover performance bottlenecks, improve software readiness, increase validation throughput, and reduce tape-out risk for advanced AI processors.