Computing & Storage

Coordinating Compute Offload, Model Data Paths and Development Resources

AI and high-performance computing platforms face growing application workloads, model-weight movement, compression, encryption and storage-reliability demands. Specialized on-chip acceleration must work effectively with system software and data-processing workflows.

Foresemi provides cryptographic and compression offload, AI weight compression and protection, and data-reliability IP for servers, AI CPUs, accelerators and storage-related chips, supporting coordinated chip and system design.

Crypto & Compression Offload

Model-Weight Compression & Protection

Compute & Data Coordination

Industry Context & Design Challenges

PCIe & CXL High-Speed Interconnects

AI accelerators, memory expanders and high-speed storage increase data exchange across chips and devices. Higher bandwidth makes security overhead, latency and interoperability important architecture-level decisions. PCIe and CXL IDE are examples of link-layer mechanisms for confidentiality, integrity and replay protection; their adoption depends on protocol versions, topology and device support.

Off-Chip Memory, Model Assets & Storage Media

Model weights and runtime data in DDR face different exposure risks from persistent data on NVMe SSDs. DDR access involves off-chip visibility, bandwidth budgets and loading latency; persistent storage also requires host authorization, media encryption and key-lifecycle management. NVMe self-encrypting storage and the host/storage authentication and authorization addressed by IEEE 1667 operate at different layers. Link encryption cannot replace either.

Protection Trade-Offs at High Throughput

Interception or tampering on a link, unauthorized reading of storage media and transmission errors are distinct problems. Evaluate link protection, key management and ECC separately, selecting implementations against bandwidth, latency and host-compute budgets. For highly concurrent workloads, cryptographic and compression offload also affects the CPU resources available to applications.

AI-Assisted Development & the Software Supply Chain

AI coding assistants can accelerate development, but generated code may contain defects, use vulnerable dependencies or resemble license-restricted third-party code. Review and statically analyze code before integration, check dependencies and licenses with software composition analysis, and retain SBOMs for released versions. Management and protocol interfaces also need malformed-input and boundary-condition testing.

Chip & System Capabilities

Cryptographic & Compression Offload

CryptACC® offloads encryption, decryption, compression and decompression from Arm and RISC-V servers and AI CPUs, supporting TLS and other secure-communication and data-processing workloads while preserving CPU resources for applications.

Model-Weight Data Paths

CryptICE™ coordinates weight compression, decompression, encryption and decryption between the DDR controller and on-chip compute units, reducing transfer volume while protecting off-chip model assets.

Storage & Data Reliability

DH-ECC provides ECC error correction and CRC checking across different data widths and correction requirements, supporting data-path protection in computing and storage platforms.

Discuss Your Solution

Share your computing platform, cryptographic and compression workloads, model-weight and data-bandwidth requirements. We will arrange a technical discussion.

Contact Us