Niobium Microsystems
Niobium Microsystems is a Dayton, Ohio-based hardware company founded in 2021 that builds a PCIe accelerator card and cloud platform purpose-built for Fully Homomorphic Encryption (FHE).
Publisher review
Niobium Microsystems is a Dayton, Ohio-based hardware company founded in 2021 that builds a PCIe accelerator card and cloud platform purpose-built for Fully Homomorphic Encryption (FHE). The company's Niobium Fog™ platform provides provable privacy through mathematically guaranteed FHE, keeping data encrypted even while in use. Its custom mistic™ hardware accelerates FHE computations, aiming to make encrypted compute practical for real-world business applications. The target audience includes enterprises handling sensitive data—such as healthcare, finance, and government—that require zero-trust computing with post-quantum security, as well as developers who need tools to build encrypted applications without deep cryptography expertise.
The core technology is BASALISC, an integrated hardware accelerator for the BGV, CKKS, and BFV FHE schemes, now in fabrication. The accelerator generates twiddle factors and key-switching keys on-chip to reduce memory bandwidth bottlenecks. Niobium's PCIe card supports all major FHE schemes and can be inserted into any server for fast integration. The company offers a developer partner program for The Fog, the first IaaS purpose-built for FHE, with SDKs, templates, and pre-built applications that allow deployment in minutes. The platform uses a zero-trust architecture where data arrives encrypted, stays encrypted, and leaves encrypted, with no exposure of plaintext data to the server.
Niobium has raised $23 million in an oversubscribed funding round (total over $28 million) from investors including Blockchange Ventures, ADVentures, Korea Development Bank, and JobsOhio Growth Capital. The company competes in the emerging FHE hardware acceleration space against general-purpose solutions like CPUs and GPUs, which are "blindingly slow" for FHE—Niobium claims a needed acceleration factor of 10^4 to 10^5 over current CPU/GPU performance. Named competitors in the broader AI hardware space include Axelera AI, Qualcomm, and SynSense, though Niobium's focus on encrypted compute is distinct from typical AI inference chips.
The honest trade-offs: FHE causes massive data expansion—ciphertexts in the 20 MB range for 20-bit plaintext precision at 128-bit security—so memory bandwidth remains a fundamental bottleneck. Even with hardware acceleration, FHE execution is still significantly slower than plaintext computation, requiring 10,000x to 100,000x speedup over current CPUs/GPUs to be practical. The programming model is unfamiliar to most developers, with complex trade-offs between security, precision, and speed. The hardware is not yet production-ready; the second-generation ASIC is still in development, and early access is limited to a developer partner program.
How it works
-
Fully Homomorphic Encryption platform
Niobium Fog provides provable privacy by keeping data encrypted during computation, enabling zero-trust architectures for sensitive workloads.
-
BASALISC hardware accelerator
A custom ASIC supporting BGV, CKKS, and BFV schemes that accelerates FHE computations, targeting 10,000x to 100,000x speedup over CPUs.
-
On-chip twiddle factor generation
Generates twiddle factors and key-switching keys on the chip, reducing memory bandwidth bottlenecks by avoiding external fetches.
-
PCIe accelerator card
A plug-in card that fits into any server, integrating FHE acceleration into existing workflows without major infrastructure changes.
-
Developer partner program
Provides SDKs, templates, and pre-built applications for The Fog IaaS, enabling deployment of encrypted applications in minutes.
-
Zero-trust data architecture
Ensures data arrives encrypted, stays encrypted, and leaves encrypted, with no plaintext exposure to the server or cloud provider.
-
Multi-scheme FHE support
Supports all major FHE schemes with post-quantum secure encryption, offering flexibility for diverse encrypted compute use cases.
Strengths and trade-offs
Strengths
- Mathematically guaranteed privacy through FHE ensures data remains encrypted even during computation, enabling zero-trust architectures.
- BASALISC hardware accelerator is purpose-built for FHE, targeting a 10,000x to 100,000x speedup over current CPU/GPU execution.
- On-chip generation of management data (twiddle factors, key-switching keys) reduces memory bandwidth bottlenecks by avoiding external memory fetches.
- Raised $23 million in an oversubscribed funding round, bringing total funding to over $28 million, to accelerate second-generation FHE hardware development.
Trade-offs
- FHE causes massive data expansion—ciphertexts in the 20 MB range for 20-bit plaintext precision at 128-bit security—creating persistent memory bandwidth challenges.
- Even with hardware acceleration, FHE execution is still significantly slower than plaintext computation, requiring 10,000x to 100,000x speedup over current CPUs/GPUs to be practical.
- The programming model is unfamiliar to most developers, with complex trade-offs between security, precision, and speed that require specialized toolkits.
- The second-generation production ASIC is still in development, and early access is limited to a developer partner program, not yet generally available.
Pricing context
Not specified in the sources; early access is available through a developer partner program with pre-built FHE applications and hardware acceleration.
Getting started with Niobium Microsystems
-
Apply to developer partner program
Visit the Niobium Microsystems website and submit an application for the developer partner program. Provide your organization details and use case to gain early access to The Fog IaaS platform and hardware acceleration for FHE.
-
Connect your server to the PCIe card
Insert the Niobium PCIe accelerator card into an available slot in your server. Ensure the server meets the hardware requirements and power specifications, then boot the system to recognize the card.
-
Configure the FHE scheme and parameters
Use the provided SDK to select the FHE scheme (BGV, CKKS, or BFV) and set security, precision, and speed parameters. Adjust ciphertext size and key-switching settings to balance performance with your application needs.
-
Load and encrypt your dataset
Use the SDK templates to load your dataset and encrypt it with the FHE keys. The data remains encrypted throughout processing, ensuring zero-trust privacy. Verify the encryption process completes without errors.
-
Run a pre-built encrypted computation
Deploy one of the pre-built FHE applications from the partner program to perform a computation on the encrypted data. Monitor the execution time and output to confirm the accelerator is functioning correctly.
Frequently Asked Questions
What is Niobium Microsystems?
Niobium Microsystems is a Dayton, Ohio-based hardware company founded in 2021 that builds a PCIe accelerator card and cloud platform for Fully Homomorphic Encryption. Its Niobium Fog platform keeps data encrypted during computation, enabling zero-trust architectures for sensitive data in healthcare, finance, and government.
How does Niobium's BASALISC hardware accelerator work?
BASALISC is a custom ASIC that accelerates FHE computations for the BGV, CKKS, and BFV schemes. It generates twiddle factors and key-switching keys on-chip to reduce memory bandwidth bottlenecks. The accelerator targets a 10,000x to 100,000x speedup over current CPU and GPU execution.
What is the Niobium Fog platform?
Niobium Fog is an Infrastructure-as-a-Service platform purpose-built for Fully Homomorphic Encryption. It provides provable privacy through mathematically guaranteed FHE, ensuring data arrives encrypted, stays encrypted, and leaves encrypted. The platform includes SDKs, templates, and pre-built applications for fast deployment.
Who is Niobium Microsystems targeting with its FHE technology?
Niobium targets enterprises handling sensitive data, such as healthcare, finance, and government, that require zero-trust computing with post-quantum security. It also serves developers who need tools to build encrypted applications without deep cryptography expertise, offering a developer partner program for early access.
What are the main trade-offs of using Fully Homomorphic Encryption?
FHE causes massive data expansion, with ciphertexts around 20 MB for 20-bit plaintext precision at 128-bit security. Even with hardware acceleration, FHE is significantly slower than plaintext computation, requiring 10,000x to 100,000x speedup over CPUs/GPUs. The programming model is also unfamiliar to most developers.
How does Niobium's PCIe card integrate into existing servers?
Niobium's PCIe accelerator card supports all major FHE schemes and can be inserted into any standard server. This allows fast integration into existing workflows without major infrastructure changes, enabling encrypted compute for sensitive workloads in a zero-trust architecture.
Alternatives
How Niobium Microsystems compares
Direct head-to-head against 3 competitors. Picked by 7wData.
Niobium Microsystems
- Pricing
- Not specified in the sources; early access is available through a developer partner program with pre-built FHE applications and hardware acceleration.
- Target
- Niobium Microsystems is a Dayton, Ohio-based hardware company founded in 2021 that builds a PCIe accelerator card and cloud platform purpose-built for Fully Homomorphic Encryption
- Strength
- Mathematically guaranteed privacy through FHE ensures data remains encrypted even during computation, enabling zero-trust architectures.
- Watch for
- FHE causes massive data expansion—ciphertexts in the 20 MB range for 20-bit plaintext precision at 128-bit security—creating persistent memory bandwidth challenges.
Optalysys
- Pricing
- Custom/Contact sales
- Target
- Secure computing with optical FHE acceleration
- Deployment
- On-premise hardware
- Strength
- Optical computing for FHE, enabling high-speed encrypted data processing
- Watch for
- Early-stage technology; limited commercial deployments and scalability
Axelera AI
- Pricing
- Custom/Contact sales
- Target
- Edge AI inference with high efficiency and low power
- Deployment
- Edge devices and modules
- Strength
- High TOPS/Watt ratio for edge inference in smart cameras and IoT
- Watch for
- Focused on inference only; no training or FHE capabilities
Qualcomm
- Pricing
- Custom/Contact sales
- Target
- Mobile and edge AI inference across smartphones, automotive, IoT
- Deployment
- Embedded SoCs and modules
- Strength
- Dominant in mobile AI with Snapdragon Neural Processing Engine
- Watch for
- Pricing tied to high-volume mobile licensing; less flexible for niche FHE
User reviews
No user reviews yet. Be the first to write one.
Sources
Reporting on this tool draws on these publicly available sources.