IN Brief:
- AMD has introduced three tamper-resistant Versal devices based on its AI Edge Series Gen 2 architecture.
- A bonded sensor die monitors specified backside physical attacks, alongside expanded cryptographic and voltage glitch detection capabilities.
- Sampling is expected in the second half of 2027, with production planned for the first half of 2028.
AMD has introduced a range of tamper-resistant Versal adaptive computing devices incorporating physical attack detection within the semiconductor package, targeting equipment required to protect sensitive information against attempts to access or manipulate the underlying hardware.
The Versal TR portfolio comprises three devices based on AMD’s AI Edge Series Gen 2 architecture. Each combines heterogeneous computing resources with a separate sensor-based protection die bonded to the main compute die through three-dimensional active-on-active integration.
The protection die covers the backside of the computing silicon and is designed to detect specified physical attacks. AMD says the architecture also prevents the device from booting if the protection die is tampered with or removed, providing an additional defence against attempts to bypass its protective circuitry.
Physical semiconductor attacks present a different challenge from vulnerabilities affecting ordinary system software. An attacker with access to electronic hardware may attempt to probe internal structures, disturb electrical operating conditions, or manipulate a device to obtain sensitive information.
These threats are relevant to embedded systems that store cryptographic keys or process information requiring protection even when an attacker can physically access the equipment. Potential applications include critical industrial infrastructure, financial systems, secure networking, aerospace equipment, and defence electronics.
Conventional protection approaches can involve specialist enclosures, monitoring circuits, secure semiconductor devices, and mechanical barriers. Their effectiveness depends on the attack methods considered and the security architecture of the completed system.
AMD’s integrated approach places additional monitoring circuitry directly against the computing die. The three-dimensional bonding arrangement creates a physical relationship between the protection layer and the silicon it is intended to defend, rather than relying entirely on circuitry located elsewhere on the system board.
The architecture is designed to identify interference with the monitored backside region and respond before normal device operation can continue. Its protective scope remains defined by AMD’s threat model; the presence of a bonded sensor die does not establish immunity against every physical attack method.
Integrating protection into the semiconductor package may also reduce the requirement for separate board-level security components in suitable applications. However, the complete system must still protect external interfaces, power connections, memory devices, communications pathways, and any other components capable of exposing sensitive information.
The Versal TR devices incorporate additional cryptographic capabilities alongside the physical protection structure. Hardware acceleration has been expanded to support the Module-Lattice-Based Digital Signature Algorithm (ML-DSA) and Stateless Hash-Based Digital Signature Algorithm (SLH-DSA).
Both algorithms are associated with post-quantum digital signature protection. Their inclusion gives developers hardware support for cryptographic mechanisms intended to withstand attack methods that could become practical as quantum computing capabilities develop.
Digital signature operations are used to verify authenticity and integrity in applications such as secure boot and software authentication. Supporting these calculations in dedicated hardware can reduce reliance on general-purpose processor resources, although the resulting performance depends on the selected algorithm and implementation.
AMD is also introducing hybrid boot support using multiple signature algorithms. Secure boot establishes whether the software or configuration being loaded has been appropriately authorised before the device proceeds into normal operation.
A hybrid arrangement can support different cryptographic mechanisms within the boot process, allowing system developers to plan transitions between existing signature methods and newer algorithms. The security of the completed implementation still depends on key management, signature verification, configuration, and the broader system design.
The devices also incorporate improvements to the application security unit key vault and a new memory element ageing sensor. These additions are intended to strengthen protection around cryptographic keys, although AMD has not published sufficient implementation detail to establish the protection mechanisms or measurable effectiveness of every new function.
Voltage glitch detection is another area of development. AMD has combined improvements to its existing architecture glitch detector with a new orthogonal detector, extending coverage to include disturbances affecting two voltage rails simultaneously.
A voltage glitch attack deliberately changes a device’s electrical operating conditions in an attempt to disturb normal execution. If precisely timed, a disturbance may interfere with a security check or cause a processor to behave incorrectly.
Monitoring abnormal supply conditions provides one mechanism for detecting such attacks, but the effectiveness depends on disturbance characteristics, detector behaviour, and the response implemented by the system. AMD has not disclosed detailed detection thresholds or independent attack resistance measurements for the new devices.
The underlying processing architecture combines Arm CPUs, AI Engines, programmable logic, dedicated hardware IP, and high-speed interfaces. Across the announced range, AMD specifies up to 229,000 DMIPS of scalar processing capability, 369 INT8 TOPS of AI computation, and 543,000 programmable logic lookup tables.
The portfolio comprises the 2VE3358TR, 2VE3558TR, and 2VE3858TR. Each has a footprint-compatible standard Versal AI Edge Series Gen 2 counterpart available for development before the tamper-resistant version enters production.
AMD says the corresponding devices retain matching computing, memory, and interface resources, allowing existing designs to migrate without repeating logic synthesis or place-and-route operations. This provides an early development route, although final implementation and qualification must still address the security characteristics of the completed hardware.
Compatibility with an existing footprint can simplify the preparation of circuit boards and development platforms. It does not remove the need to verify electrical operation, firmware behaviour, system integration, and application-specific protection requirements when the new devices become available.
The distinction between the currently available standard devices and the future TR portfolio is important. AMD expects the first tamper-resistant samples in the second half of 2027, followed by production in the first half of 2028. The announcement does not establish that production versions of the protected devices are shipping today.
AMD also states that tamper resistance should not be interpreted as tamper-proof operation. The protection applies to specified physical attack methods and must be evaluated against the intended security threat model.
The Versal TR portfolio introduces a defined combination of backside monitoring, cryptographic acceleration, and glitch detection within a programmable computing architecture. Its practical contribution to security-critical equipment will depend on device qualification and the performance of the complete protective system once production hardware becomes available.



