Access Official Technical PDF

Share Summary:LinkedInEmail

Silicon prototyping can help computer architecture teams test ideas in hardware rather than relying solely on simulation. Historically, however, chip development has demanded specialized tools, extensive engineering effort and substantial funding. The BRGTC2 research chip from Cornell University illustrates how open-source infrastructure, productive design tools and multi-project wafer (MPW) services can make the process more approachable for small teams.

Why silicon prototyping has been difficult

A chip project spans much more than writing RTL. Teams may need to model an architecture, develop and test hardware, synthesize the design, plan its floorplan, complete physical implementation and verify that the resulting layout meets manufacturing requirements. These stages depend on one another, and moving from a working design to tapeout can take considerable time and coordination.

Closed tools and intellectual property can add friction. If key parts of a design or toolchain are difficult to inspect, reproduce or maintain, teams may struggle to share results and build on prior work. Limitations in one layer can also affect others, from the instruction set architecture and software to RTL development and the ASIC flow. For research groups, this can mean spending effort recreating infrastructure rather than evaluating the architectural question at hand.

RISC-V connects an open design ecosystem

The RISC-V ecosystem offers an open instruction set architecture around which researchers can develop software, model systems and build hardware. In the BRGTC2 project, upstream GCC support and RISC-V modularity supported incremental design work. The team also used gem5 for multicore, cycle-level simulations to explore design decisions before committing to an implementation.

Open RTL projects and related infrastructure—including Rocket, BOOM, PULP, Diplomacy, FIRRTL and FireSim—form additional parts of the ecosystem described in the project. Community reference ASIC methodologies can also give teams a starting point for implementation. These resources do not eliminate the need for project-specific design and verification, but they can reduce the amount of foundational infrastructure a team must create on its own.

For BRGTC2, simulation helped evaluate questions such as how many resources to share, the effect of resource latencies, the use of L0 buffers and whether target programs would fit in the cache. This kind of exploration lets architects compare options at the cycle level before investing in detailed RTL and physical design.

Productive tools and modular ASIC flows

Small teams face the same broad sequence of design and verification tasks as larger organizations, but with fewer people available to manage tools and flow maintenance. A rigid ASIC flow can make experimentation harder: inserting or removing a step may require changes to the build system, while separate chip and block configurations can lead to duplicated, project-specific flows.

The BRGTC2 work used PyMTL, a Python-based hardware modeling framework for design and testing, along with an open modular VLSI build system. The goal of a modular flow is to treat implementation stages as components that can be assembled as needed. Individual steps can manage their own execution and cleanup, while the build system checks prerequisites and outputs.

This approach allows a flow to be assembled at different levels of completeness. A team might use a smaller graph for an architecture study, include additional steps for a VLSI evaluation, or assemble a fuller flow for tapeout. The practical benefit is flexibility: the same underlying build system can support different project needs without treating every new design as an entirely separate flow.

BRGTC2 also used a fully synthesizable, all-digital phase-locked loop (PLL). The project described this as a way to avoid mixed-signal design for that component. Its use illustrates how design choices and supporting tools can affect the complexity of integrating a chip, though it does not remove the broader verification and implementation work.

BRGTC2: a compact RISC-V research chip

BRGTC2, or Batten Research Group Test Chip 2, was fabricated in TSMC 28 nm and measured 1 mm by 1.25 mm. The reported design contained 6.7 million transistors and four in-order RISC-V RV32IMAF cores. It included shared 32 KB L1 instruction and data caches, shared functional units, a hardware bloom filter accelerator and the synthesizable PLL. The design was developed and tested using PyMTL and ran a work-stealing runtime.

The project also shows that chip development is a team activity extending beyond processor RTL. The reported work included application development, porting the runtime to the RISC-V target, gem5 design-space exploration, component and SRAM testing, integration testing at RTL and gate level, PLL modeling, I/O floorplanning, and physical design and post-placement-and-routing tuning.

MPW services and the economics of prototyping

Fabrication has traditionally been a major barrier to experimental silicon. MPW services address this by allowing multiple designs to share a wafer run, making small production quantities possible without requiring a team to fund an entire wafer run for its own design. The whitepaper identifies access to small minimum sizes in advanced nodes as an important change for research prototyping.

For BRGTC2, the reported fabrication cost was about $18,000 for a 1 mm by 1.25 mm die and 100 parts through the MOSIS Tiny2 program. Packaging was about $2,000 for 20 parts, while PCB and assembly costs were reported at less than $1,000. These historical project figures are not a current price quote; total project costs also included student salaries, physical IP and EDA tool licenses.

What engineering teams can take from the example

The BRGTC2 experience suggests several practical considerations for teams assessing a silicon prototype:

  • Build on open infrastructure where it fits. An open ISA, software tools, simulators and RTL resources can support reproducibility and reduce dependence on closed components.
  • Explore architecture before implementation. Cycle-level simulation can help teams examine resource sharing, latency and workload fit before detailed hardware work.
  • Keep implementation flows adaptable. Modular steps make it easier to tailor a flow to a research question, an evaluation or a tapeout.
  • Plan for the whole project. Integration, verification, floorplanning, packaging, boards and tool or IP costs all matter alongside core design work.

Conclusion

Open RISC-V resources, productive hardware modeling and modular ASIC flows address important sources of complexity in silicon prototyping. MPW services can make small fabrication runs more accessible, although a prototype still requires coordinated effort across design, verification and physical implementation. For industrial automation engineers and technical decision-makers evaluating custom silicon or architecture research, the central lesson is to assess the complete development ecosystem—not just the chip design—and plan tools, people and fabrication together.

Access Official Technical PDF

About The Author