sky130ram: the macro LEFs leave met4 and all vias out of OBS (and part of met1–met3), so routes to the pins end up too close to macro metal

已关闭
  • 哪里不对？
#4,753 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
48/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
领域
backend, databases

调研方向

Start by comparing the sky130ram macro LEFs with their GDS shapes using the KLayout script described in the issue; inspect PIN and OBS RECT/POLYGON coverage on met1–met4 and via layers. The issue names the four affected macros and reports that rebuilding OBS from GDS cleared the reproduced DRC markers. Done means the routing and cut shapes outside signal pins are represented in OBS and the reproduced design has no such markers.

由索引模型根据 Issue 内容生成。

描述

Description

Comparing each sky130ram macro's GDS with its LEF (PIN and OBS shapes, RECT and POLYGON both counted), metal that is in the GDS but in neither PIN nor OBS:

macro met1 met2 met3 met4
sky130_sram_1rw1r_44x64_8 1,018 µm² (8.0 %) 646 (5.1 %) 898 (13.1 %) all 98 shapes, 418 µm²
sky130_sram_1rw1r_64x256_8 3,230 (5.1 %) 1,771 (2.5 %) 5,882 (10.2 %) all 144 shapes, 1,235 µm²
sky130_sram_1rw1r_80x64_8 1,320 (6.1 %) 948 (4.1 %) 1,718 (9.6 %) all 177 shapes, 819 µm²
sky130_sram_1rw1r_128x256_8 5,063 (4.1 %) 2,640 (1.9 %) 5,297 (4.9 %) all 288 shapes, 1,890 µm²

The LEFs have no met4 at all (the met4 shapes sit along the edge with the signal pins) and no cut layers in OBS. The missing met1/met2 includes long lines, e.g. met1 151.485..151.745 × 79.545..585.445 in the 128x256 macro.

The router takes that space as free. In a small test design (a 128 × 256 FIFO on one sky130_sram_1rw1r_128x256_8, sky130hd, ORFS c06bf3c; routing DRC 0), sky130hd.lydrc found 466 markers that the macro checked on its own does not have, all inside the macro's footprint next to its pins: m1.2 148, m2.2 185, m3.2 40, m4.2 72, via.2 8, m2.7 6, m1.1 4, m1.7 3 — e.g. a met4 wire to a pin 0.135 µm from the macro's own met4 line at 332.915..333.215 × 3.305..31.05, and dout0 met1 wires 0.105 µm from unlisted met1. With an OBS rebuilt from the GDS (union of the old OBS and all met1–met4/via–via3 shapes, keeping a 0.5 µm way in to each signal pin on its own metal) the same design routes clean and the count drops to 0. (Separately from these, the macro alone has 320,604 markers from the missing SRAM core-area handling, #4735.)

To reproduce

The comparison needs only the platform files; a KLayout script that reads the LEF's PIN and OBS RECT/POLYGON shapes into one region per layer and subtracts it from the GDS layer (68/20, 69/20, 70/20, 71/20) gives the table above. The files are unchanged on current master (2021-02 OpenRAM output); no design in flow/designs/ uses them.

Suggested fix

Regenerate the LEFs (or the macros, see the vdd/gnd issue #4752) so OBS covers every routing-layer and cut-layer shape outside the pins; or document the gap and ship a script that rebuilds OBS from the GDS.

主要语言
Verilog
星标
782
派生
552
平均合并
2 天 17 小时
30 天内合并 PR
73

环境准备

  • 提供 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 没有贡献指南

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

The-OpenROAD-Project/OpenROAD-flow-scripts 的其他 Issue

查看 The-OpenROAD-Project/OpenROAD-flow-scripts 的全部 Issue

相似的 Issue

更多 Backend & API Design Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。