Hardware-enforced security

Software will make mistakes. Limit what happens next.

CHERI gives processors a stronger way to control memory access, helping systems block invalid operations and contain compromised components.

Understand the problem

01 / The problem

The problem reaches everyday life.

Software sometimes accesses memory it should not. One mistake can expose personal information, disrupt a service or let an attacker change part of a device. Memory safety is a human, commercial and resilience issue.

  • At homeConnected devices

    Privacy, availability and long product lifetimes depend on trustworthy software boundaries.

  • On the moveVehicles and transport

    Safety-critical functions coexist with connectivity, third-party code and demanding updates.

  • In careMedical equipment

    Reliability, controlled access and clear fault detection remain valuable for years.

  • In publicCritical services

    A compromised component should not automatically gain authority over a whole system.

Read everyday examples

02 / Where risk appears

Complex software sits inside familiar products.

Phones, routers, connected appliances, vehicles, medical equipment and public infrastructure all depend on software. Memory errors do not always become attacks, but exploitable errors can expose data, corrupt another component or interrupt a service.

Keep the claim proportionate. A case example should identify the affected system and evidence. It should not imply that every incident was caused by memory unsafety.

03 / Defence in depth

Good engineering matters. It needs layers.

Testing, analysis, memory-safe languages and conventional mitigations all do important work. CHERI adds hardware-enforced restrictions. It is an additional layer, not a replacement.

  1. Layer 01Reduce exploitability

    Use conventional mitigations and secure configuration.

  2. Layer 02Find mistakes

    Use review, testing, fuzzing, analysis and sanitizers.

  3. Layer 03Prevent mistakes

    Use memory-safe languages and carefully designed interfaces.

  4. Layer 04Enforce authority

    Use CHERI bounds, permissions, validity and compartments.

04 / A long-standing idea

Authority, with limits.

Capability systems are not a sudden invention. CHERI builds on an idea developed and tested over decades, adapted for modern processors and software.

  1. 1960s–70sCapability foundations

    Early systems establish authority carried by protected references.

  2. 2010CHERI begins

    Cambridge and SRI combine capability ideas with modern processor architecture.

  3. 2014–20Hardware and software

    Architecture, FPGA processors, CheriBSD and formal methods advance together.

  4. 2022 onwardLarge-scale evaluation

    Morello and a growing RISC-V and embedded ecosystem expand practical evaluation.

Follow the complete history

05 / What CHERI changes

Every pointer carries authority the hardware can check.

Capabilities carry information used to enforce where software may access memory and what it may do there.

  1. Step 01Attempt

    Software attempts to access a memory address.

  2. Step 02Validate

    The processor checks protected validity, bounds and permissions.

  3. Step 03Decide

    A valid request continues; an invalid request faults at the point of use.

  4. Step 04Contain

    A compartment boundary can limit the authority of a compromised component.

Explore the technology

06 / Validation

From foundations to whole-system evidence.

CHERI grew through sustained hardware, software and formal-methods development. Research programmes and development platforms enabled progressively broader evaluation.

Dates, platform status and measured outcomes should link to primary sources. Separate architectural properties from results that depend on implementation and system design.

Follow the evidence

07 / Ecosystem

From architecture to working systems.

CHERI spans processors, boards, operating systems, compilers, languages and tools. Every entry needs a clear maturity label and a dated source.

  • Development platformArm Morello

    Large-scale CHERI evaluation across a rich operating-system stack.

  • Open FPGA platformSonata

    Embedded research, teaching and prototyping.

  • Embedded architectureCHERIoT

    Fine-grained compartments for embedded systems.

  • Software stackCheriBSD and LLVM

    Operating-system and compiler support for application evaluation.

Compare platform categories

08 / Outcomes, qualified

Change the path from mistake to impact.

Hardware-enforced restrictions can reduce opportunities to exploit memory errors, improve fault detection and support stronger isolation.

  1. 01Software mistake

    The underlying programming error may still occur.

  2. 02Restricted authority

    The component has only the access it needs.

  3. 03Invalid access blocked

    The processor checks the request at the point of use.

  4. 04Compromise contained

    A compartment boundary may limit wider impact.

Important: CHERI does not remove programming mistakes or make systems unhackable. Outcomes depend on implementation, integration and wider system design.

09 / The economic decision

Security is a lifecycle decision.

Compare investigation, patching, downtime and long-term support with the real costs of hardware change, porting, training, testing and assurance.

  • BaselineCurrent exposure

    Support, incident, recall, downtime and assurance costs.

  • AdoptionTransition cost

    Hardware, porting, tools, skills, testing and certification.

  • ComparisonExplicit ranges

    Low, central and high assumptions with evidence quality shown.

Use the text-only business-case worksheet

10 / Complementary controls

CHERI and Rust are stronger together.

Memory-safe languages prevent many mistakes in new code. CHERI adds hardware-enforced protection across mixed-language systems and established code.

  • Language layerRust and memory-safe languages

    Prevent broad classes of errors and express ownership and lifetimes.

  • Hardware layerCHERI capabilities

    Enforce bounds and permissions across new, mixed and established code.

  • System outcomeLayered memory safety

    Combine prevention, enforcement and stronger containment.

Compare the approaches

11 / Why an Alliance

Adoption is a team sport.

An architecture is not enough. Practical adoption needs platforms, maintained tools, shared evidence, skills, assurance and interoperable specifications.

  • Hardware and softwareInteroperable systems

    Coordinate architecture, platforms, tools and maintained software.

  • Research and educationEvidence and skills

    Make learning, evidence and implementation guidance visible.

  • Industry and governmentAdoption and assurance

    Connect practical needs to standards, policy and procurement.

Learn about the Alliance