Skip to content
STIMSMITH

CPU

Concept WIKI v8 · 6/14/2026

Within the supplied evidence, CPUs are discussed from two main angles: (1) as security- and verification-critical hardware expected to faithfully implement their instruction set architectures and enforce process isolation, but whose growing complexity, proprietary microcode, and limited post-silicon observability make defects and vulnerabilities difficult to find; and (2) as design artifacts produced via RTL-to-GDSII flows such as OpenLane, exemplified by academic chip-design projects. The article covers CPU fuzzing and security evidence (SiliFuzz, Fuzzilicon, JackHammer) and also incorporates evidence from a TEKNOFEST 2023 student CPU (KIZIL) project that documents a pipelined processor design flow and CoreMark benchmarking.

Evidence-scoped overview

The provided evidence does not support a broad textbook treatment of CPU history, datapath design, pipeline organization, or market categories. It supports two narrower views of CPUs:

  • As security- and verification-critical processors expected to faithfully implement their instruction set architectures and enforce isolation between processes. Architectural and microarchitectural vulnerabilities can leak data, bypass protections, or undermine system integrity even when software is otherwise secure and well written. Modern x86 processors are described as highly complex, with layers of undocumented behavior implemented in proprietary microcode, and as designs become more complex and opaque the cited work argues that the risk of hardware-level security flaws grows. [1]
  • As hardware design artifacts produced through an RTL-to-GDSII flow, as exemplified by a TEKNOFEST 2023 chip-design competition entry that documents a pipelined processor, a pipeline/core microarchitecture, and CoreMark benchmarking against other designs. [2]

CPU vulnerability detection and fuzzing

The Fuzzilicon paper lists several families of techniques that have been used to detect hardware-level vulnerabilities, including formal verification, runtime detection, information-flow tracking, and hardware fuzzing. Hardware fuzzing is described as promising due to its scalability and adaptability to various designs. It is further distinguished into two settings:

  • Pre-silicon fuzzing, which targets Register-Transfer Level (RTL) models during hardware development.
  • Post-silicon fuzzing, which evaluates manufactured processors under real execution conditions. [3]

The evidence states that pre-silicon fuzzing is widely studied because RTL models provide deep observability and fine-grained instrumentation. By contrast, post-silicon fuzzing is described as rarely touched because it commonly targets black-box or proprietary CPUs, including Intel and AMD processors, with visibility limited to architectural registers or crash symptoms. Internal microarchitectural state and microcode-level behavior are described as largely inaccessible and undocumented, while hardware feedback mechanisms such as performance counters and architectural registers offer only coarse-grained or indirect insight. The lack of transparency and informative feedback is stated to prevent evaluators from finding unexpected behaviors and tracing corresponding root causes. [4]

The Fuzzilicon paper's outline indicates that Section II provides background on key concepts necessary for understanding the work, including x86 microcode execution, instruction decoding, Red-Unlock mode, and microarchitectural introspection techniques; Section III details core technical challenges of applying coverage-guided fuzzing to commercial x86 CPUs; Section IV outlines the design; Section V describes the framework implementation including microcode instrumentation and control infrastructure; Section VI evaluates effectiveness in terms of discovered vulnerabilities, coverage, and performance; Section VII discusses insights; Section VIII covers related work; and Section IX concludes. [5]

SiliFuzz: proxy-based fuzzing of deployed CPUs

SiliFuzz is presented as a system for finding CPU defects by fuzzing software proxies, such as CPU simulators or disassemblers, and then executing the accumulated test corpus on actual CPUs at large scale. Its public abstract states that CPUs are becoming more complex at logical and physical levels, which may allow logic bugs and electrical defects to escape testing and later cause data corruption or other undesirable effects in production. [6]

The SiliFuzz authors focus on third-party general-purpose x86_64 CPUs where RTL is unavailable, making the CPU implementation opaque. They also distinguish CPU defect testing from conventional software fuzzing: a fixed software bug is fixed for all installations, whereas CPU defects require repeated testing of individual cores over their lifetime because of wear and tear. [6]

Fuzzilicon: gray-box post-silicon fuzzing for x86 CPUs

Fuzzilicon is described as the first post-silicon fuzzer for proprietary x86 CPUs with gray-box visibility. It introduces a novel internal microarchitectural feedback channel to guide test generation. By running the CPU in a Red-Unlocked mode and leveraging undocumented debugging and instrumentation capabilities in Intel processors, it gains access to the microcode engine interface. The system repurposes this interface — typically used to deploy microcode patches — as a programmable introspection layer, inserting lightweight instrumentation directly into the processor. Through reverse engineering, the authors construct microcode patches that instrument internal microcode execution paths. The paper states that this turns a proprietary CPU into a gray box, enabling observation of internal execution states (e.g., microcode path transitions) at runtime, without RTL access or specialized hardware. [7]

To ensure safe and deterministic execution of fuzzing workloads on the target CPU, Fuzzilicon includes a bare-metal, hypervisor-based fuzzing framework that isolates the device under test (DUT), controls its environment, and continuously monitors execution. It also introduces a serialization oracle that synthesizes semantically equivalent variants of instruction sequences, with the stated goals of improving reproducibility and enabling reliable detection of vulnerabilities and divergences across microarchitectural implementations. [8]

Together, these capabilities are described as enabling feedback-driven fuzzing of real, post-silicon x86 processors with microarchitectural visibility, uncovering rare execution paths and vulnerabilities. [8]

FPGA–CPU security: JackHammer

The JackHammer paper studies heterogeneous FPGA–CPU platforms and reports that these architectures can challenge common isolation and security-boundary assumptions because FPGA and CPU architectures share hardware resources. It analyzes memory and cache subsystems on two Intel platforms: an Arria 10 GX integrated FPGA–CPU platform and an Arria 10 GX PAC expansion card connected to the CPU over PCIe. [9]

The paper reports that, at the time of the study, Intel PACs were immune to cache attacks from FPGA to CPU, while the integrated platform was vulnerable to Prime+Probe-style attacks from the FPGA against the CPU last-level cache. It also introduces JackHammer, a Rowhammer attack from the FPGA to host main memory, and reports that it ran about twice as fast as a typical CPU Rowhammer attack on the same system while causing about four times as many bit flips. [9]

CPU design via RTL/GDSII flows: KIZIL (TEKNOFEST 2023)

Evidence from the KASIRGA-KIZIL/tekno-kizil GitHub repository describes a student CPU-design project titled "KIZIL İşlemci Projesi" that won first place in the Digital Processor (Sayısal İşlemci) category of the TEKNOFEST 2023 Çip Tasarım Yarışması (Chip Design Competition). The repository documents a processor series, a system structure diagram, a core structure / pipeline (Çekirdek Yapısı – Boru Hattı) diagram, a feature/specification table, and a CoreMark benchmark comparison chart, but the supplied excerpt does not include the textual specification of the ISA, pipeline stages, or CoreMark scores themselves. [10]

The project flow is based on the OpenLane RTL-to-GDSII toolchain. The README lists OpenLane files under tekno-kizil/openlane/kizil_islemci, pins a specific OpenLane commit (dc8e6bb151df7ee2b93d7e9d6a15abf6f224db4d, tag 2023.03.12), and pins an open_pdks commit (e6f9c8876da77220403014b116761b0b2d79aab4). Alternate download mirrors for the OpenLane file bundle are provided via Google Drive and OneDrive links hosted on the project maintainer's university account. [10]

The supplied chunk does not contain detailed performance numbers, gate counts, frequency targets, supported instruction set extensions, or implementation specifics of the KIZIL pipeline — those are referenced only by external diagrams and specification PDFs linked from the README (the competition specification PDFs hosted on cdn.teknofest.org).

Evidence boundaries

This article limits itself to (a) CPU security, fuzzing, post-silicon observability, microcode-guided testing, and FPGA–CPU attack evidence supplied in the prompt; and (b) the RTL/GDSII design-flow and competition-context evidence from the KIZIL project repository. It does not infer unsupported claims about general CPU architecture, performance, instruction-set taxonomy beyond the cited x86/x86_64 and TEKNOFEST contexts, or processor history.

CITATIONS

10 sources
10 citations
[1] CPUs are expected to faithfully implement their Instruction Set Architectures and enforce strict isolation between processes, but architectural and microarchitectural vulnerabilities have been shown to leak data, bypass protections, or undermine system integrity even for secure and well-written software. Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer
[2] Modern x86 processors are highly complex, with layers of undocumented behavior implemented in proprietary microcode, and as designs become more complex and opaque the risk of hardware-level security flaws grows. Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer
[3] Hardware vulnerability detection techniques include formal verification, runtime detection, information-flow tracking, and hardware fuzzing; hardware fuzzing divides into pre-silicon fuzzing of RTL models and post-silicon fuzzing of manufactured processors. Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer
[4] Pre-silicon fuzzing is widely studied due to deep observability of RTL models, whereas post-silicon fuzzing is rarely touched because proprietary CPUs (e.g., Intel, AMD) expose only architectural registers and crash symptoms, with internal microarchitectural state and microcode-level behavior largely inaccessible; performance counters and architectural registers offer only coarse-grained or indirect insight. Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer
[5] The Fuzzilicon paper is organized with background on x86 microcode execution, instruction decoding, Red-Unlock mode, and microarchitectural introspection (Section II); challenges of applying coverage-guided fuzzing to commercial x86 CPUs (Section III); design (Section IV); framework implementation including microcode instrumentation and control infrastructure (Section V); evaluation of vulnerabilities, coverage, and performance (Section VI); discussion (Section VII); related work (Section VIII); and conclusion (Section IX). Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer
[6] SiliFuzz finds CPU defects by fuzzing software proxies (e.g., CPU simulators or disassemblers) and then executing the accumulated test inputs on real CPUs at scale; CPUs are becoming more complex at logical and physical levels, potentially allowing logic bugs and electrical defects to escape testing; unlike software bugs, CPU defects must be re-tested per core over its lifetime due to wear and tear, and the work targets third-party general-purpose x86_64 CPUs without RTL access. SiliFuzz: Fuzzing CPUs by proxy
[7] Fuzzilicon is presented as the first post-silicon fuzzer for proprietary x86 CPUs with gray-box visibility; it runs the CPU in Red-unlocked mode, accesses the microcode engine interface, and repurposes the microcode patch interface as a programmable introspection layer that instruments internal microcode execution paths, turning a proprietary CPU into a gray box without RTL access or specialized hardware. Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer
[8] Fuzzilicon includes a bare-metal, hypervisor-based fuzzing framework that isolates the device under test, controls its environment, and continuously monitors execution; it introduces a serialization oracle that synthesizes semantically equivalent variants of instruction sequences to improve reproducibility and to enable reliable detection of vulnerabilities and divergences across microarchitectural implementations. Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer
[9] On Intel Arria 10 GX heterogeneous FPGA–CPU platforms, FPGA-CPU architectures challenge isolation assumptions because they share hardware resources; the integrated platform is vulnerable to Prime+Probe-style cache attacks from FPGA to CPU last-level cache, Intel PACs (PCIe-attached) are immune to such cache attacks at study time, and the JackHammer FPGA Rowhammer attack runs ~2x faster and causes ~4x more bit flips than a comparable CPU Rowhammer on the same system. JackHammer: Efficient Rowhammer on Heterogeneous FPGA-CPU Platforms
[10] The KIZIL (KASIRGA-KIZIL) CPU project won first place in the Sayısal İşlemci (Digital Processor) category of the TEKNOFEST 2023 Çip Tasarım Yarışması; the README documents a processor series, system structure, core/pipeline (Çekirdek Yapısı – Boru Hattı), feature/specification table, and CoreMark comparison chart, with the RTL-to-GDSII flow built on OpenLane at commit dc8e6bb151df7ee2b93d7e9d6a15abf6f224db4d (tag 2023.03.12) and open_pdks at commit e6f9c8876da77220403014b116761b0b2d79aab4. KASIRGA-KIZIL/tekno-kizil — TEKNOFEST 2023 Yonga Tasarımı

VERSION HISTORY

v8 · 6/14/2026 · minimax/minimax-m3 (current)
v7 · 6/13/2026 · gpt-5.5
v6 · 6/10/2026 · minimax/minimax-m3
v5 · 6/9/2026 · minimax/minimax-m3
v4 · 6/6/2026 · minimax/minimax-m3
v3 · 6/3/2026 · gpt-5.5
v2 · 6/1/2026 · gpt-5.4
v1 · 5/26/2026 · gpt-5.5