TDEB Top Bar

Trending
Semiconductor Industry News — The Digital Electronics Blog
Follow Us

TDEB Masthead

The Semiconductor Hub

Semiconductors.  Chips and Systems.  Innovation.

Indexed by leading Search Engines, AI and referenced by Engineers worldwide 500+ Articles 10M+ Pageviews 30+ Reports 50+ Citations

Special Report

Why Is IP Integration So Hard? Interfaces, Power, Test and Chiplets Explained

|Report

By Murugavel Ganesan

Murugavel Ganesan
by
0
Share Facebook X LinkedIn WhatsApp Email
A bare system-on-chip die under studio light, with memory arrays, logic regions, I/O PHY blocks and an analog block visible as distinct areas.

Almost no chip today is designed from scratch. A modern SoC is assembled from processor cores, interconnect, memory controllers, PHYs, security blocks and analog functions, many of them bought from vendors or reused from earlier projects. Each block arrives verified on its own. The trouble starts when they are wired together, and that step, IP integration, is where a large share of schedule slips and silicon bugs now come from.

This special report explains why integration is hard and what engineering teams do about it. It starts with a plain-words overview, then covers the details from protocols and power intent to chiplets and metadata standards. The starting point is Brian Bailey's Nov 12, 2026 article in Semiconductor Engineering, and every fact links to the source it came from.

Key takeaways

  • Most integration bugs live between blocks, not inside them. Each IP can pass its own verification while the system hangs because two teams read the same protocol rule differently.
  • The volume has grown. Engineers quoted by Semiconductor Engineering (Nov 12, 2026) describe a move from tens of IP blocks per design a decade ago to thousands of IPs per company today, with dozens of versions of the same block.
  • The hard part is often missing information: incomplete specifications, undocumented firmware rules and untested configurations.
  • Standards carry much of the load: IEEE 1685 (IP-XACT) for metadata, IEEE 1801 (UPF) for power intent, IEEE 1735 for encrypted IP, Accellera's new CDC/RDC standard (Mar 2, 2026) for clock and reset crossings, and UCIe for die-to-die links.
  • Chiplets move the integration boundary from the chip to the package, which turns IP integration into a problem that spans companies and foundries.

In this post

What IP integration is

In chip design, "IP" means a reusable block of design: a CPU core, a USB controller, a DDR memory PHY, a crypto engine, a PLL. Some IP comes as source code (soft IP) that is synthesized with the rest of the chip. Some comes as a finished layout (hard IP), common for analog and high-speed blocks. Either way, the block is designed and verified by one team, often at another company, and then dropped into a chip by a different team.

Integration is everything that happens after the drop. The integrator connects the block's buses to the on-chip interconnect, assigns it addresses, wires its interrupts, gives it clocks and resets, puts it in the right power domain, hooks up its test logic, places it on the floorplan, closes timing around it, and writes or adapts the software that drives it. Then the whole chip has to be verified as a system.

Why it is hard

Every IP block carries assumptions about the world around it. A simple UART needs a clock, a reset and a register interface. A coherent cache controller or a network-on-chip carries hundreds of assumptions about ordering, credits, outstanding transactions, reset and power sequencing, address maps and security, as Axiomise founder Ashish Darbari explains in the Semiconductor Engineering article. Most of those assumptions are written down somewhere, but not always completely, and not always in a form the integrator reads the same way the IP designer meant it.

Darbari's summary is blunt: "The nastiest failures are almost never inside a single block." His examples include deadlock that depends on message order, leaked credits, corrupted reorder buffers and ordering rules broken across bridges.

Three other forces make it worse. First, scale: Keysight's Pedro Pires notes in the same article that a design that once had tens of IP blocks now sits in a company that tracks thousands of IPs and dozens of versions of each. Second, software: even a PHY now ships with boot and calibration firmware, and the hardware often works only if that software follows rules nobody wrote down. Third, power: blocks now change power state while their neighbors are talking to them.

The cost of getting it wrong is high. The 2024 Wilson Research Group study for Siemens EDA found that only 14% of IC and ASIC projects achieved first-silicon success, the lowest level the study has recorded in two decades, according to the trend report by Harry Foster. The study does not single out integration, but it shows how little room there is for error.

The main problem areas

  • Interfaces. Two blocks that both "speak AXI" may still disagree on widths, ordering or optional features.
  • Configuration. Configurable IP can be built thousands of ways, and the vendor tested only some of them.
  • Clocks, resets and power. Signals crossing clock, reset or power domains need special handling, and each block's assumptions must carry up to the chip.
  • Test and security. Every block's test logic must be reachable from the chip pins, and encrypted IP hides how a block really behaves.
  • Physical design. Hard blocks have fixed shapes, pins and power needs that drive the floorplan and timing.
  • Chiplets and tracking. Dies from different companies must work as one system, and someone has to know which version of each block and its firmware went into which chip.

What teams do about it

Write things down in a machine-readable form. The biggest gains come from replacing PDFs and spreadsheets with formats tools can check. IP-XACT describes a block's interfaces, registers and files. UPF describes its power intent. Accellera's new CDC/RDC standard describes its clock and reset crossings. When the IP vendor ships these, the integrator's tools can check the connection rather than a person reading a datasheet.

Check connections automatically. Formal tools can prove that every wire in the SoC goes where a connectivity table says it should, which is faster and more complete than simulating it. Verification IP (prebuilt protocol checkers and models) catches protocol violations at each interface.

Choose and qualify IP carefully. Foundry programs such as TSMC9000 score IP on design reviews and silicon results. Teams also ask vendors which configurations were tested and in which silicon.

Track everything. Record which version of each block, and which firmware, went into which tape-out.

None of this removes the problem. Pires says the tools improve but the problem grows, so the pain stays about constant in relative terms.

Deep dive: the numbers and details

This part holds the standards, numbers, a worked example and a checklist. Skip it if the overview is all you need.

Interface and protocol mismatch

Standard buses solved the first-order problem. Arm describes AMBA as an open, royalty-free standard for connecting blocks in an SoC, usable with any processor architecture, and lists a family of protocols: APB for low-bandwidth register traffic, AHB mostly for Cortex-M class embedded designs, AXI for high-bandwidth interconnect, and CHI for fully coherent processors, with CHI keeping separate protocol and transport layers (Arm, AMBA page).

A shared protocol name does not guarantee a working connection. The places where integration breaks are the ones the protocol leaves open or optional:

  • Widths and ordering. Different data, address and ID widths mean bridges and converters, each a place where ordering can change, and credits and buffers must be sized for the worst case.
  • Coherency. Every agent must agree on which transactions snoop and how memory attributes are set.
  • Address maps, interrupts and QoS. Synopsys' Madhumita Sanyal lists address mapping, interrupts, coherency, QoS and security as the main cross-IP issues at SoC level in the Semiconductor Engineering article.

The deeper problem is semantic. Darbari's point is that architects, designers, software and verification teams can each read the same failure differently, because intended behavior was never written in one shared form, and the gaps often surface only in post-silicon debug.

Software adds another layer: hardware often works only if software programs registers in a set order or avoids writes during in-flight transactions, rules Darbari says are rarely documented. For more on on-chip fabrics, see our explainer on network-on-chip IP.

Configuration explosion and parameter checking

Much IP is now delivered as a generator or a heavily parameterized RTL package. That is good for reuse and bad for verification coverage. Darbari points out in the Semiconductor Engineering article that with generator-based IP the vendor may have verified different configurations from the one you are building, and that datasheets often do not explain what each parameter does to behavior.

Worked example (illustrative numbers). Take a hypothetical interconnect bridge IP with these build parameters. The parameters are invented to show the arithmetic.

Parameter (illustrative)Allowed valuesCount
Data width32, 64, 128, 256, 512 bits5
Address width32, 40, 48, 52 bits4
ID width4 to 16 bits13
Maximum outstanding transactions8, 16, 32, 644
ECC on data pathon, off2
Clockingsynchronous, asynchronous2
Retention in low-power modeon, off2

The number of distinct builds is the product of the counts: 5 x 4 x 13 x 4 x 2 x 2 x 2 = 4,160. Suppose, again as an illustration, the vendor runs its full regression on 40 configurations. That is 40 / 4,160, or under 1% of the space.

What to do with that number:

  1. Ask the vendor for the list of configurations in its regression and in silicon, and pick from it where you can.
  2. Where you cannot, rerun the vendor's test suite on your exact configuration.
  3. Check parameters at the boundary. A width or ID mismatch between a master and a slave should fail an automated check, not a simulation three months later. IP-XACT lets parameters and their legal ranges be described in metadata (see below).

Clocks, resets, CDC and RDC across IP

Each IP may have several clocks and resets of its own. When a signal crosses between asynchronous clocks, or between logic reset by different resets, it can go metastable, glitch or reconverge in the wrong state. Clock domain crossing (CDC) and reset domain crossing (RDC) checks find these paths. See our post on CDC verification.

The integration problem is collateral. An IP vendor runs CDC and RDC analysis with its own tool and ships the results, or an abstract model of the block, to the integrator. Until this year, those results could not be reused in another vendor's tool, so integrators reran analysis and translated results by hand. Accellera's Board approved the Standard for IP Abstraction for Clock and Reset Domain Crossing Integration 1.0 on Mar 2, 2026, defining a vendor-neutral abstraction of a block's CDC/RDC intent that sits alongside HDL and IP-XACT descriptions (Accellera, Mar 2, 2026).

The integrator still checks, at each IP boundary, that every expected clock and reset is supplied with the right relationship, that declared synchronizers sit on the real crossing paths, and that power-up reset ordering matches the IP's assumptions.

Power intent across IP boundaries

Power management has become a functional feature. Darbari's example in the Semiconductor Engineering article is a block that works when powered but fails when a neighbor powers down in the middle of a transfer. Every power domain adds states, transitions, isolation and wake-up sequences, and he argues that such cases are formal problems because simulation rarely hits the exact cycle.

The standard here is IEEE 1801, the Unified Power Format. IEEE 1801-2024, board-approved Sep 26, 2024 and published Mar 4, 2025, defines how to specify a design's power intent and verify the design against it, and it supports incremental refinement of power intent for IP-based flows (IEEE SA, IEEE 1801-2024). Incremental refinement is the integration feature: an IP vendor writes the block's own power intent, and the integrator adds the chip-level supplies and strategies without rewriting it.

Hard macros need extra care. A 2020 Siemens (then Mentor) article explains that a hard macro is usually delivered as a behavioral model or netlist for verification and a Liberty cell for implementation, so the parent design sees it as a black box whose ports are domain boundaries. Its conclusion is that hard macros need a self-contained UPF covering boundary supplies, internal power states and how those states map to system power states (Progyna Khondkar, Semiconductor Engineering, May 14, 2020). Without it, simulation may assume the macro is powered from the wrong supply and miss isolation errors.

Agile Analog's Chris Morrison adds in the Bailey article that analog IP now helps manage power and heat, with embedded sensors, timing monitors and droop and glitch detectors sending telemetry to the power management unit.

DFT and test integration

Each IP brings its own test structures: scan chains, memory BIST, PHY loopback, sometimes its own JTAG port. A chip with dozens of such blocks cannot give each one its own pins. A 2014 Synopsys article in Electronic Design describes the standard answer, IEEE 1500: a wrapper around each core with an instruction register (WIR) to select test modes, boundary registers (WBR) to control and observe the core's edges, and a bypass (WBY), all reached through serial and parallel test access mechanisms and described in Core Test Language (Prasad Saggurti, Electronic Design, Jul 17, 2014).

The article's integration advice: chain wrappers in a ring rather than a tree, expose one 1500 interface per level of hierarchy, reuse IP-level patterns at chip level, schedule tests to trade test time against power and pin count, and disable wrapper access after manufacturing test.

For access to embedded instruments such as sensors and BIST controllers, IEEE 1687 (IJTAG), published Dec 5, 2014, defined an on-chip network from the instruments to the chip pins through the IEEE 1149.1 test access port, plus languages to describe that network and to talk to the instruments. The IEEE lists the 2014 edition as inactive-reserved since Mar 27, 2025 (IEEE SA, IEEE 1687-2014). See our DFT interview questions post for the basics.

Security and encrypted IP

Vendors protect their RTL by encrypting it. IEEE 1735-2023, published Nov 6, 2023, is the recommended practice for this: it covers encryption of design IP, the usage rights granted by the IP producer, management of those rights, license checks, markup that works with SystemVerilog and VHDL, the choice of encryption algorithms and key management, and the trust model it assumes (IEEE SA, IEEE 1735-2023).

Darbari observes in the Semiconductor Engineering article that you can no longer read the RTL to learn how a block really behaves, so you depend on documentation, which is where bugs often hide.

The second security issue is what each block assumes about its neighbors: which bus masters may reach its registers, whether debug access is locked, what happens to secrets on reset. Accellera's Security Annotation for Electronic Design Integration (SA-EDI) 1.0, approved Jul 14, 2021, gives IP providers a uniform way to document security concerns in their IP so they either mitigate them or disclose them to the integrator (Accellera release via Design & Reuse). Accellera contributed it to IEEE as the basis of draft standard P3164 in January 2023 (Design & Reuse, Jan 23, 2023).

Verifying the integrated chip

Verification takes 60 to 70% of engineering effort in chip projects, according to Harry Foster of Siemens EDA (Verification Horizons, Sep 3, 2025). For integration, the work splits into three layers.

Connectivity. The first question is whether every wire goes where it should: interrupts, clocks, resets, test signals, pin multiplexing, debug buses. Formal connectivity apps take a specification, often a spreadsheet table that lists each connection with its gating conditions and register delays, and prove the RTL matches it. A 2019 sponsored article in Semiconductor Engineering argues that simulation exercises "only a tiny fraction of possible behavior" for this kind of check, while formal covers all of it and prunes unrelated logic so it runs on a full chip (Tom Anderson, Jan 31, 2019).

Protocols. Verification IP (VIP) supplies protocol checkers and traffic models; a checker on every IP boundary catches interface bugs early.

System behavior. The remaining bugs (deadlock, ordering, power sequencing, firmware assumptions) need system tests, software running on the model or an emulator, and targeted formal proofs. Our post on formal verification explains the method. For analog IP, Morrison notes that SPICE is too slow at chip scale, so teams wrap analog blocks in SystemVerilog real-number models for full-chip digital simulation.

Physical integration

Hard IP sets fixed constraints on the floorplan. A Sondrel article on integrating SerDes, DDR and PCIe IP into a large 5 nm SoC gives a practical picture (Subindrao Johal and Pavan Patel, Design & Reuse):

  • Collateral early. The authors stress getting IP collateral and the deliverables list as early as possible; incomplete LEF files make block-level and chip-level LVS difficult.
  • Power hookup. Where hard IP power ports do not line up with standard-cell power stripes, manual connections are needed.
  • ECO loops. PT-ECO signoff on large blocks typically needed more than twelve iterations because of noise and transition limits.
  • Pin checks. One port placement check reported 427 violations, mostly pins off the block edge (276) or off track (145).

Physical integration depends on the quality and timing of deliverables: LEF, Liberty, GDS, power and ESD rules.

Chiplet-era integration

Chiplets move the integration boundary from the die to the package. Synopsys' Frank Schirrmeister puts it in the Bailey article as IP integration turning into supply chain integration across companies, nodes, foundries and verification methods. Sanyal lists what die-to-die IP adds: interoperability, known-good die, signal and power integrity, power delivery, thermal coupling and telemetry.

UCIe is the main open standard for the die-to-die link. The UCIe specifications page lists optional manageability, a DFx architecture for test and debug, and 3D packaging support in UCIe 2.0, and an engineering change notice adding chiplet security, lifecycle state reporting and test and debug access control. UCIe 3.0, announced Aug 5, 2025, raised rates to 48 and 64 GT/s, standardized early firmware download through the Management Transport Protocol, added runtime recalibration, extended sideband reach to 100 mm, and added fast throttle and emergency shutdown signals (UCIe Consortium via Design & Reuse). UCIe 4.0 adds UCIe.mem for memory connectivity, an optional power-optimized profile that Electronic Design reports gives 30% to 80% better energy efficiency at 64 GT/s, secure two-way management channels, and Trusted Reporting of Chiplet State (William G. Wong, Electronic Design, Nov 16, 2026).

A working link is not a working system. Darbari notes in the Semiconductor Engineering article that UCIe handles the physical and link layers well and cross-vendor links have been shown, but address maps, coherency and software agreements across dies remain open. Arm's CHI C2C, which packetizes CHI traffic for chip-to-chip links, is one attempt to carry coherency across that boundary (Arm, AMBA page). Our post on UCIe in practice covers the link itself.

IP quality, qualification and versioning

TSMC describes TSMC9000 as a baseline of IP quality requirements with staged assessments for hard IP: a pre-silicon review of design kits and margins, a test chip tape-out review for major IPs, and typical and split-lot silicon assessments from test chip characterization. Scores are posted for customers and IPs with known issues are labeled. TSMC also states it does not guarantee that silicon-verified IP will work in every design environment (TSMC, TSMC9000 page).

Versioning is the other half. IC Manage's Dean Drako says in the Semiconductor Engineering article that teams must track what was taped out, built and used where, and which IP went into which design. Pires adds that links between hardware and firmware versions often live in spreadsheets, and argues for a traceable system bill of materials. Darbari warns that field firmware updates can change silicon behavior after sign-off.

Metadata: IP-XACT and related formats

IEEE 1685-2022, IP-XACT, approved Sep 21, 2022 and published Feb 28, 2023, standardizes XML descriptions of components and systems, bus interfaces and their connections, bus abstractions, address maps, registers and fields, and file sets, with schemas, semantic consistency rules and a generator interface that works across tools (IEEE SA, IEEE 1685-2022). Accellera hosts the schema files (current version dated Feb 13, 2025), conversion scripts, an example and a user guide (Accellera, IP-XACT downloads).

For registers specifically, Accellera's SystemRDL 2.0 lets one register description generate RTL, documentation and other outputs that stay consistent; the current release dates from Jan 26, 2018, and the working group is dormant (Accellera, SystemRDL downloads).

Metadata is only as good as the vendor's discipline in keeping it in step with the RTL, so check it against the RTL on delivery.

Problem areas at a glance

AreaTypical failureCollateral or standard that helpsCheck that catches it
InterfacesWidth, ID or ordering mismatch; deadlock across bridgesAMBA specs, IP-XACT bus interfacesProtocol VIP, formal deadlock and ordering proofs
ConfigurationUntested parameter combinationIP-XACT parameters, vendor regression listRerun vendor tests on your configuration
Clocks and resetsMissing synchronizer, reset ordering errorAccellera CDC/RDC standard 1.0CDC and RDC analysis with IP abstracts
PowerIsolation missing, neighbor powers down mid-transferIEEE 1801 (UPF), self-contained macro UPFPower-aware simulation, formal on power sequences
TestCore test unreachable, test time too longIEEE 1500, CTL, IEEE 1687Pattern retargeting and test scheduling
SecurityUnlocked debug, unchecked bus accessIEEE 1735, SA-EDISecurity review against IP disclosures
ConnectivityInterrupt or pin-mux wired wrongConnectivity table, IP-XACT designFormal connectivity check
PhysicalIncomplete LEF, power hookup, ECO loopsLEF, Liberty, GDS, ESD rulesEarly LVS, DRC and timing on IP views
ChipletsLink up, system broken (address map, coherency)UCIe, CHI C2CMulti-die system simulation and emulation
TrackingWrong IP or firmware version in a tape-outIP management system, system bill of materialsRelease audit before tape-out

A practical checklist

  1. Get the deliverables list in writing: RTL, IP-XACT, UPF, CDC/RDC abstract, test collateral, LEF, Liberty, GDS, firmware and security disclosures.
  2. Confirm every tool in your flow can read the IEEE 1735 encrypted RTL.
  3. Pick configurations the vendor has regressed or taped out, and rerun its tests on yours.
  4. Check foundry qualification status and the known-issues list for your node.
  5. Generate wiring, register maps and headers from metadata, not by hand.
  6. Put a protocol checker on every IP interface and run formal connectivity on interrupts, clocks, resets, test and pin multiplexing.
  7. Run chip-level CDC and RDC with the IP abstracts, and review every waiver.
  8. Run power-aware simulation with each IP's UPF, including a neighbor powering down mid-transfer.
  9. Write down software rules such as register programming order, and test them with real firmware.
  10. Get physical views early and run LVS and DRC on them.
  11. Before tape-out, record every IP version, configuration and firmware version in a system bill of materials, and confirm debug access is locked.
  12. For chiplets, agree address maps, coherency rules and management firmware with every die supplier in writing.

What to watch

  • Part two of the Semiconductor Engineering series. The Nov 12, 2026 article says a second part will cover recent improvements and the use of AI in IP selection, configuration and integration (Semiconductor Engineering).
  • Tool support for the CDC/RDC standard. Accellera approved version 1.0 on Mar 2, 2026 (Accellera). Watch for IP shipping with abstracts in this format.
  • UCIe 4.0 adoption. Electronic Design reported the UCIe 4.0 features on Nov 16, 2026 (Electronic Design). Watch for first silicon using UCIe.mem.
  • IEEE P3164. The IEEE version of Accellera's security annotation standard has been in development since its kick-off in February 2023 (Design & Reuse).
  • The next first-silicon success figure. The 2024 Wilson Research Group study put it at 14% (Verification Academy). The next edition of the study will show whether the trend has turned.

Sources

  • Brian Bailey, "IP Integration Challenges Evolve", Semiconductor Engineering, Nov 12, 2026 (primary reference): semiengineering.com
  • Harry Foster, "IC/ASIC Functional Verification Trend Report - 2024", Siemens EDA and Wilson Research Group, paper dated Feb 17, 2025: verificationacademy.com
  • Harry Foster, "Why First-Silicon Success Is Getting Harder for System Companies", Verification Horizons blog, Sep 3, 2025: blogs.sw.siemens.com
  • Arm, "AMBA" architecture page: arm.com
  • IEEE SA, "IEEE 1685-2022, IEEE Standard for IP-XACT, Standard Structure for Packaging, Integrating, and Reusing IP within Tool Flows", published Feb 28, 2023: standards.ieee.org
  • Accellera, "Download IP-XACT" (schemas, user guide, examples): accellera.org
  • Accellera, "Download SystemRDL" (SystemRDL 2.0): accellera.org
  • IEEE SA, "IEEE 1801-2024, IEEE Standard for Design and Verification of Low-Power, Energy-Aware Electronic Systems", published Mar 4, 2025: standards.ieee.org
  • Progyna Khondkar, "Power Management And Integration Of IPs In SoCs: Part 2", Semiconductor Engineering (sponsored), May 14, 2020: semiengineering.com
  • IEEE SA, "IEEE 1735-2023, IEEE Recommended Practice for Encryption and Management of Electronic Design Intellectual Property (IP)", published Nov 6, 2023: standards.ieee.org
  • Accellera, "Accellera Approves Clock and Reset Domain Crossing (CDC/RDC) Standard 1.0 for Release", Mar 2, 2026: accellera.org
  • Accellera, "Accellera Board Approves Security Annotation for Electronic Design Integration Standard 1.0 for Release", Jul 14, 2021, copy on Design & Reuse: design-reuse.com
  • Accellera, "Accellera's Security Annotation for Electronic Design Integration Standard 1.0 Moves Toward IEEE Standardization", Jan 23, 2023, copy on Design & Reuse: design-reuse.com
  • Prasad Saggurti, "Apply IEEE 1500 to Integrate Multiple JTAG Chains in SoCs", Electronic Design, Jul 17, 2014: electronicdesign.com
  • IEEE SA, "IEEE 1687-2014, IEEE Standard for Access and Control of Instrumentation Embedded within a Semiconductor Device", published Dec 5, 2014: standards.ieee.org
  • Tom Anderson, "Connectivity Checking Is A Perfect Fit For Formal Verification", Semiconductor Engineering (sponsored), Jan 31, 2019: semiengineering.com
  • Subindrao Johal and Pavan Patel (Sondrel), "Integrating high speed IP at 5nm", Design & Reuse (undated): design-reuse.com
  • UCIe Consortium, "Specifications" page: uciexpress.org
  • UCIe Consortium, "UCIe Consortium Introduces 3.0 Specification With 64 GT/s Performance and Enhanced Manageability", Aug 5, 2025, copy on Design & Reuse: design-reuse.com
  • William G. Wong, "UCIe 4.0 Standardizes Memory Connectivity", Electronic Design, Nov 16, 2026: electronicdesign.com
  • TSMC, TSMC9000 IP quality program page: tsmc.com

Views attributed to named engineers are as reported in the Semiconductor Engineering article cited above. Standards details are as published by IEEE, Accellera, Arm and the UCIe Consortium on the dates shown. The worked example uses illustrative numbers that do not describe any real product.

Murugavel Ganesan
About the Author
Murugavel Ganesan
Product Director | SerDes · IVN · Edge AI · Compute | Ex Intel · Infineon

A semiconductor product leader working across product architecture, definition, strategy, and silicon and software execution, from concept through volume production. He has worked with teams across Asia, Europe and the USA, including at Intel, Infineon and ArrayComm.

His work covers ASICs, SoCs, FPGAs and chiplets for wireless, wireline, automotive, IoT and AI, including PCIe, CXL and DDR interfaces alongside RF and mixed-signal design.

Share Facebook X LinkedIn WhatsApp Email

Post a Comment

0 Comments

Your comments will be moderated before it appears here.

Post a Comment (0)

#buttons=(Ok, Go it!) #days=(20)

Our website uses cookies to enhance your experience. Learn more
Ok, Go it!