IN Brief:
- Encrypted computing is moving from software prototypes to dedicated accelerators in mainstream nodes.
- Niobium is targeting fully homomorphic encryption throughput with a purpose-built ASIC on Samsung 8nm.
- The programme links design services, foundry manufacture, and downstream packaging and test for deployment-ready hardware.
Niobium has taken a step towards volume manufacture of its fully homomorphic encryption (FHE) accelerator ASIC, selecting Samsung Foundry as the manufacturing partner and SEMIFIVE as the design and delivery partner for the chip.
FHE has long been treated as the “nice idea, impossible runtime” corner of cryptography: it enables computation directly on encrypted data, so workloads can be processed without exposing plaintext at any point in the pipeline. The catch is cost. Even well-optimised software stacks can turn routine inference, analytics, or database operations into heavyweight, latency-prone runs once the arithmetic is lifted into an encrypted domain, pushing engineers towards specialised acceleration.
The new programme targets that bottleneck with dedicated silicon on Samsung Foundry’s 8nm Low Power Ultimate (8LPU) process. SEMIFIVE’s scope spans more than front-end design work, covering packaging and testing, alongside supply chain management intended to shorten the path from architecture to manufacturable parts.
Kevin Yoder, CEO of Niobium, said: “Encrypted computation will become inevitable. Once enterprises can compute directly on encrypted data at fast enough speeds, processing sensitive information in the clear will no longer be acceptable. With SEMIFIVE and Samsung Foundry, we’re translating years of R&D into production-ready silicon for encrypted cloud and AI environments. This represents the next step in our transition from prototype systems to accelerators suitable for customer deployments.”
The push aligns with wider movement in “privacy-first” compute, where regulated data, model inputs, and intermediate results are increasingly expected to remain protected end-to-end. In practice, that spans confidential analytics, private inference, and collaborative computation across organisational boundaries where raw data cannot be shared, but outcomes still need to be computed with predictable latency and cost.
Brandon Cho, CEO and co-founder of SEMIFIVE, positioned the design win as a high-complexity ASIC delivery play, saying: “Niobium is at the forefront of encrypted computation, and their FHE accelerator platform represents one of the most important emerging architectures in privacy-first computing. We are pleased to support Niobium in translating this cutting-edge FHE innovation into manufacturable silicon through SEMIFIVE’s proven SoC development and platform execution.”
Samsung, meanwhile, is framing the work inside its foundry ecosystem, with Taejoong Song, vice president and head of Foundry Technology Planning at Samsung Electronics, stating: “Encrypted computation will play an increasingly critical role in the future of AI and cloud systems. Through Samsung Foundry’s advanced process technology and SAFE™ ecosystem of partners, we are proud to support SEMIFIVE and Niobium as they bring next-generation private computing silicon to global markets.”
For system designers, the near-term question is less about whether FHE is “possible” and more about where accelerator silicon can land on the performance-per-watt curve, and how cleanly it can be integrated into existing cloud and AI infrastructure stacks. Moving to an 8nm production flow is a sign that the target is no longer academic feasibility, but repeatable deployment.
