AiEDA/iDATA

Dataset

iDATA is a dataset of AI + EDA, which can be used to train AI models for chip design PPA prediction, PPA-aware physical design, and related tasks.

Dataset structure

Describe the dataset structure.

aes/
├── syn_netlist/    # The synthesized netlist files、sdc files
├── place/          # The place stage def、sdc、vectors
└── route/          # The route stage def、sdc、vectors
    ├── rpt/        # The spef files
    ├── vectors/    # Stage-specific feature data
    ├── aes_a_route_congestion_best.def      # Routed def
    ├── aes_a_route_congestion_best.v        # Routed verilog
    └── ...    # more routed def/sdc

The def and verilog file name contain:

• 'a' meaning that the synthesized netlist is used area effort strategy.
• 'b' meaning that the synthesized netlist is used to ppa balance strategy.
• 'n' meaning that the synthesized netlist is used to no optimize strategy.
• 'p' meaning that the synthesized netlist is used to power effort strategy.
• 't' meaning that the synthesized netlist is used to timing effort strategy.
• 'congestion_best' meaning the place stage def is the best one with the lowest congestion.
• 'congestion_worst' meaning the place stage def is the worst one with the highest congestion.
• 'HPWL_best' meaning the place stage def is the best one with the lowest HPWL.'
• 'HPWL_worst' meaning the place stage def is the worst one with the highest HPWL.

The vectors/ directory contains feature data for each strategy design.

Dataset creation

Source data

The source data is collected from oscc-ip, iscas89, OpenRoad, and OpenLane.

Data processing

The data processing uses the AiEDA toolkit.The following steps show how to use the toolkit to process the data:

• SDC preprocessing: run python AiEDA/test/process_sdc_files.py --dataset-dir <dataset_dir> to automatically generate placement and routing SDC files with/without set_propagated_clock [all_clocks] (written to place/ and route/, respectively).
• Batch placement runs: execute python AiEDA/test/test_sky130_place.py --design-list <json> (add --summary-out, --first-class-only, etc. if needed). The script drives iEDA according to the workspace, Verilog, SDC, and SPEF entries in the design list to perform floorplan→route; outputs are saved in each workspace under output/iEDA/. The design list is an array where each item must include at least design, workspace, verilog, and sdc, and may optionally include spef, clk_port_name, etc. Example:

  [
    {
      "design": "gcd",
      "workspace": "/path/to/workspaces",
      "verilog": "/path/to/syn_netlist/gcd.v",
      "sdc": "/path/to/syn_netlist/gcd_place.sdc",
      "spef": "",
      "clk_port_name": ["clk"]
    }
  ]
  

• Batch extract placement features: run python AiEDA/test/batch_extract_place_features.py --dataset-dir <dataset_dir> [--patch-row-step N --patch-col-step M]. For each design, this script calls DataGeneration.generate_vectors(..., is_placement_mode=True) on place/*.def files and writes results to output/iEDA/vectors/place.
• Batch extract routing features: run python AiEDA/test/batch_extract_route_features.py --dataset-dir <dataset_dir> [--patch-row-step N --patch-col-step M]. This reads DEF/SDC/SPEF from the routing stage and generates features under output/iEDA/vectors/route.

Dataset Description

Foundation Data: Net

• Example File Path: vectors/nets/net_0.json

• Data Generation Mechanism:

  • The Vectorization::buildFeature function is called to invoke buildLayoutData and buildGraphData to construct the foundational data.
  • generateFeature is called to generate net features, including timing features, DRC features, and statistical features.
  • Finally, the data is saved via _data_manager.saveData, where Net data is saved to the nets directory using the VecLayoutFileIO::saveJsonNets method.
  • Batch processing mode is supported; depending on the batch_mode parameter, a single file can save one or multiple nets.

• Data Structure Example:

  {
    "id": 0,
    "name": "Net Name",
    "feature": {
      "llx": 79680,
      "lly": 68880,
      "urx": 82320,
      "ury": 74880,
      "wire_len": 11040,
      "via_num": 6,
      "drc_num": 0,
      "drc_type": [],
      "R": 19.71428571428571,
      "C": 0.0034665075873600007,
      "power": 1.1747088838172315e-07,
      "delay": 0.029491537147148576,
      "slew": 0.009644028041781348,
      "aspect_ratio": 2,
      "width": 2640,
      "height": 6000,
      "area": 15840000,
      "volume": 31680000,
      "layer_ratio": [0.0, 0.0, 0.2800000011920929, ...],
      "place_feature": {
        "pin_num": 2,
        "aspect_ratio": 2,
        "width": 2195,
        "height": 3507,
        "area": 7697865,
        "l_ness": 1.0,
        "hpwl": 5702,
        "rsmt": 5702
      }
    },
    "pin_num": 2,
    "pins": [
      {
        "id": 0,
        "i": "Cell Instance Name",
        "p": "Pin Name",
        "driver": 1
      }
    ],
    "wire_num": 11,
    "wires": [
      {
        "id": 0,
        "feature": {
                "wire_width": 140,
                "wire_len": 960,
                "wire_density": null,
                "drc_num": 0,
                "R": 0.8571428571428577,
                "C": 4.6356008480000124e-05,
                "power": 8.727133642824888e-07,
                "delay": 0.0022083051501257146,
                "slew": 6.621690001873404e-11,
                "congestion": 4.5,
                "drc_type": []
        },
        "wire": {
                "id1": 742163,
                "x1": 81360,
                "y1": 74880,
                "real_x1": 81360,
                "real_y1": 75055,
                "r1": 312,
                "c1": 339,
                "l1": 2,
                "p1": -1,
                "id2": 1316755,
                "x2": 81360,
                "y2": 74880,
                "real_x2": 81360,
                "real_y2": 75055,
                "r2": 312,
                "c2": 339,
                "l2": 4,
                "p2": -1,
                "via": 5
          },
        "path_num": 1,
        "paths": [
          {
                    "id1": 742163,
                    "x1": 81360,
                    "y1": 74880,
                    "real_x1": 81360,
                    "real_y1": 75055,
                    "r1": 312,
                    "c1": 339,
                    "l1": 2,
                    "p1": -1,
                    "id2": 1316755,
                    "x2": 81360,
                    "y2": 74880,
                    "real_x2": 81360,
                    "real_y2": 75055,
                    "r2": 312,
                    "c2": 339,
                    "l2": 4,
                    "p2": -1,
                    "via": 5
          }
        ],
        "patch_num": 2,
        "patchs": [528, 529]
      }
    ],
    "routing_graph": {
      "vertices": [
            {
                "id": 0,
                "is_pin": 0,
                "is_driver_pin": 0,
                "x": 81360,
                "y": 75055,
                "layer_id": 2
            }
      ],
      "edges": [
            {
                "source_id": 0,
                "target_id": 1,
                "path": [
                    {
                        "x": 81360,
                        "y": 75055,
                        "layer_id": 2
                    },
                    {
                        "x": 82320,
                        "y": 75055,
                        "layer_id": 2
                    }
                ]
            }
      ]
    }
  }
  

• Field Descriptions:

  Field Name	Data Type	Description	Unit	Example Value
  id	Integer	Unique identifier for the net	-	0
  name	String	Name of the net	-	"ctrl$a_mux_sel[0]"
  feature	Object	Collection of net features	-	-
  feature.llx	Integer	Lower-left x-coordinate of the net's bounding box, taking the leftmost polygon value	Database Unit	79680
  feature.lly	Integer	Lower-left y-coordinate of the net's bounding box, taking the bottommost polygon value	Database Unit	68880
  feature.urx	Integer	Upper-right x-coordinate of the net's bounding box, taking the rightmost polygon value	Database Unit	82320
  feature.ury	Integer	Upper-right y-coordinate of the net's bounding box, taking the topmost polygon value	Database Unit	74880
  feature.wire_len	Integer	Total length of the net's wire segments	Database Unit	11040
  feature.via_num	Integer	Number of vias	Count	6
  feature.drc_num	Integer	Number of DRC violations	Count	0
  feature.drc_type	Array	List of DRC violation types	-	[]
  feature.R	Float	Total resistance of the net	Ohms	19.714
  feature.C	Float	Total capacitance of the net	pF	0.003467
  feature.power	Float	Total power consumption of the net	Watts	1.1747e-07
  feature.delay	Float	Total delay of the net	ns	0.02949
  feature.slew	Float	Signal slew rate	ns	0.009644
  feature.aspect_ratio	Float	Aspect ratio of the bounding box	-	2.0
  feature.width	Integer	Width of the bounding box	Database Unit	2640
  feature.height	Integer	Height of the bounding box	Database Unit	6000
  feature.area	Integer	Area of the bounding box	Square DB Unit	15840000
  feature.volume	Integer	3D volume of the bounding box	Cubic DB Unit	31680000
  feature.layer_ratio	Array	Wire length distribution ratio for each layer from bottom to top	-	[0.0, 0.0, 0.28, ...]
  feature.place_feature	Object	Features from the placement stage	-	-
  feature.place_feature.pin_num	Integer	Number of pins connected to the net	Count	2
  feature.place_feature.aspect_ratio	Float	Aspect ratio of the net's bounding box, calculated from pin center points	-	2.0
  feature.place_feature.width	Integer	Width of the pin placement bounding box, calculated from pin center points	Database Unit	2195
  feature.place_feature.height	Integer	Height of the pin placement bounding box, calculated from pin center points	Database Unit	3507
  feature.place_feature.area	Integer	Area of the pin placement bounding box, calculated from pin center points	Square DB Unit	7697865
  feature.place_feature.l_ness	Float	Pin distribution, from A.B. Kahang's paper "What the L"	-	1.0
  feature.place_feature.hpwl	Integer	Half-perimeter wire length, calculated from pin center points	Database Unit	5702
  feature.place_feature.rsmt	Integer	Rectilinear Steiner minimal tree length, calculated from pin center points	Database Unit	5702
  pin_num	Integer	Number of pins	Count	2
  pins	Array	List of pins	-	-
  pins[].id	Integer	Pin ID	-	0
  pins[].i	String	Name of the instance the pin belongs to	-	"ctrl/34"
  pins[].p	String	Pin name	-	"X"
  pins[].driver	Integer	Is it a driver pin? (1: Yes, 0: No)	-	1
  wire_num	Integer	Number of wire segments	Count	11
  wires	Array	List of wire segments	-	-
  wires[].id	Integer	Wire segment ID	-	0
  wires[].feature	Object	Wire segment features	-	-
  wires[].feature.wire_width	Integer	Wire segment width	Database Unit	140
  wires[].feature.wire_len	Integer	Wire segment length	Database Unit	960
  wires[].feature.wire_density	Float	Wire segment density	-	0.1
  wires[].feature.drc_num	Integer	Number of DRC violations on the segment	Count	0
  wires[].feature.R	Float	Resistance of the segment	Ohms	0.8571
  wires[].feature.C	Float	Capacitance of the segment	pF	4.6356e-05
  wires[].feature.power	Float	Power consumption of the segment	Watts	8.7271e-07
  wires[].feature.delay	Float	Delay of the segment	ns	0.002208
  wires[].feature.slew	Float	Slew rate of the segment	ns	6.6217e-11
  wires[].feature.congestion	Float	Congestion in the wire segment's area (demand-supply)	-	4.5
  wires[].feature.drc_type	Array	List of DRC violation types on the segment	-	[]
  wires[].wire	Object	Node info for the wire segment; connection is not necessarily Manhattan	-	-
  wires[].wire.id1	Integer	Start node ID	-	742163
  wires[].wire.x1	Integer	Start x-coordinate on the grid	Database Unit	81360
  wires[].wire.y1	Integer	Start y-coordinate on the grid	Database Unit	74880
  wires[].wire.real_x1	Integer	Actual start x-coordinate	Database Unit	81360
  wires[].wire.real_y1	Integer	Actual start y-coordinate	Database Unit	75055
  wires[].wire.r1	Integer	Row number of the start point	-	312
  wires[].wire.c1	Integer	Column number of the start point	-	339
  wires[].wire.l1	Integer	Metal layer of the start point	-	2
  wires[].wire.p1	Integer	Corresponding real pin number for the start point, -1 if not a real pin	-	-1
  wires[].wire.id2	Integer	End node ID	-	742167
  wires[].wire.x2	Integer	End x-coordinate on the grid	Database Unit	82320
  wires[].wire.y2	Integer	End y-coordinate on the grid	Database Unit	74880
  wires[].wire.real_x2	Integer	Actual end x-coordinate	Database Unit	82320
  wires[].wire.real_y2	Integer	Actual end y-coordinate	Database Unit	75055
  wires[].wire.r2	Integer	Row number of the end point	-	312
  wires[].wire.c2	Integer	Column number of the end point	-	343
  wires[].wire.l2	Integer	Metal layer of the end point	-	2
  wires[].wire.p2	Integer	Corresponding real pin number for the end point, -1 if not a real pin	-	-1
  wires[].path_num	Integer	Number of Manhattan paths the wire is decomposed into	Count	1
  wires[].paths	Array	Information on the Manhattan paths the wire is decomposed into	-	-
  wires[].patch_num	Integer	Number of patches passed through	Count	2
  wires[].patchs	Array	List of IDs of patches passed through	-	[528, 529]
  routing_graph	Object	Routing graph structure, representing the net's graph theory model	-	-
  routing_graph.vertices	Array	List of vertices in the graph, containing all nodes	-	-
  routing_graph.vertices[].id	Integer	Unique vertex identifier	-	0
  routing_graph.vertices[].is_pin	Integer	Is it a pin? (1: Yes, 0: No)	-	0
  routing_graph.vertices[].is_driver_pin	Integer	Is it a driver pin? (1: Yes, 0: No)	-	0
  routing_graph.vertices[].x	Integer	Vertex x-coordinate	Database Unit	81360
  routing_graph.vertices[].y	Integer	Vertex y-coordinate	Database Unit	75055
  routing_graph.vertices[].layer_id	Integer	Metal layer ID of the vertex	-	2
  routing_graph.edges	Array	List of edges, representing connections between vertices	-	-
  routing_graph.edges[].source_id	Integer	Source vertex ID of the edge	-	0
  routing_graph.edges[].target_id	Integer	Target vertex ID of the edge	-	1
  routing_graph.edges[].path	Array	List of path points on the edge, representing the actual route	-	-
  routing_graph.edges[].path[].x	Integer	Path point x-coordinate	Database Unit	81360
  routing_graph.edges[].path[].y	Integer	Path point y-coordinate	Database Unit	75055
  routing_graph.edges[].path[].layer_id	Integer	Metal layer ID of the path point	-	2

• Data Usage:

  1. Provides the geometric and electrical characteristics of nets for physical design analysis.
  2. Supports timing analysis with key parameters like resistance, capacitance, delay, and slew.
  3. Used for DRC checks and violation analysis.
  4. Power estimation and optimization.
  5. Routing density and congestion analysis.
  6. Provides training data for machine learning models to predict net performance.

• Data Characteristics:

  1. Contains multi-level net information, from macroscopic bounding boxes to microscopic wire segment details.
  2. Integrates geometric, electrical, and timing characteristics.
  3. Supports feature analysis for both placement and routing stages.
  4. Stored in JSON format for easy parsing and processing.
  5. Provides a complete routing graph structure, facilitating graph-based analysis and algorithm optimization.

Foundation Data: Path

• Example File Path: vectors/wire_paths/wire_path_1.json

• Data Generation Mechanism:

  • First, foundational data is built using buildLayoutData and buildGraphData.
  • The buildFeatureTiming method of the VecFeature class is called for timing feature extraction.
  • eval_tp->runVecSTA() is executed to perform a detailed timing analysis.
  • Specifically, the reportWirePaths() function generates detailed path data by traversing paths in the timing graph.
  • Finally, the data is saved in JSON format to the wire_paths directory through file I/O operations.

• Data Structure Example: Path data uses a JSON array format, organizing the nodes and delay arcs of a path in chronological order. Each element in the array is a JSON object representing an element in the path (a node or an arc).

  [
    {
      "node_0": {
        "Point": "dpath/b_reg/_147_:CLK (sky130_fd_sc_hs__dfxtp_1)",
        "Capacitance": 0.003,
        "slew": 0.0,
        "trans_type": "rise"
      }
    },
    {
      "inst_arc_0": {
        "Incr": 0.238
      }
    },
    {
      "node_1": {
          "Point": "dpath/b_reg/_147_:Q (sky130_fd_sc_hs__dfxtp_1)",
          "Capacitance": 0.018471882955040004,
          "slew": 0.134413,
          "trans_type": "rise"
      }
    },
    {
      "net_arc_1": {
          "Incr": 0.000476,
          "edge_0": {
              "wire_from_node": "dpath/b_reg/_147_:Q",
              "wire_to_node": "dpath/a_lt_b$in1[15]:847926",
              "wire_R": 0.0,
              "wire_C": 0.0,
              "from_slew": 0.134413,
              "to_slew": 0.134413,
              "wire_delay": 0.0
          },
          // more edges...
      }
    }
    // more nodes and arcs...
  ]
  

• Field Descriptions:

  Field Type	Field Name	Data Type	Description	Unit	Example Value
  Node Info	node_*	Object	An instance pin or internal node on the path	-	-
  	node_*.Point	String	Node location info, format: "instance_name:pin_name (cell_type)"	-	"dpath/b_reg/147:CLK (sky130_fd_sc_hs__dfxtp_1)"
  	node_*.Capacitance	Float	Node capacitance value	pF	0.003
  	node_*.slew	Float	Node signal transition time (slew)	ns	0.134413
  	node_*.trans_type	String	Signal transition type	-	"rise" or "fall"
  Instance Arc	inst_arc_*	Object	Internal delay arc of an instance, representing cell-internal propagation delay	-	-
  	inst_arc_*.Incr	Float	Incremental delay within the instance	ns	0.238
  Net Arc	net_arc_*	Object	Net delay arc, representing signal propagation delay on the net	-	-
  	net_arc_*.Incr	Float	Total incremental delay of the net	ns	0.000476
  	net_arc_*.edge_*	Object	Specific wire segment information within the net	-	-
  	net_arc_*.edge_*.wire_from_node	String	Start node name of the wire segment	-	"dpath/b_reg/147:Q"
  	net_arc_*.edge_*.wire_to_node	String	End node name of the wire segment	-	"dpath/a_lt_b$in1[15]:847926"
  	net_arc_*.edge_*.wire_R	Float	Resistance value of the wire segment	Ω	0.642857
  	net_arc_*.edge_*.wire_C	Float	Capacitance value of the wire segment	pF	3.618685e-05
  	net_arc_*.edge_*.from_slew	Float	Slew at the start of the wire segment	ns	0.134413
  	net_arc_*.edge_*.to_slew	Float	Slew at the end of the wire segment	ns	0.134413033744
  	net_arc_*.edge_*.wire_delay	Float	Delay introduced by the wire segment	ns	1.18515e-05

• Data Usage:

  1. Timing Analysis: Provides detailed path delay information for analyzing and optimizing the design's timing performance.
  2. Signal Integrity Assessment: Includes signal transition times and net RC parameters, supporting signal integrity analysis.
  3. Power Analysis: Based on node capacitance and toggle rates, dynamic power can be calculated.
  4. Path Visualization: The data structure facilitates the graphical representation and analysis of paths.
  5. Timing Closure Verification: Used to verify that the design meets timing constraints.
  6. Machine Learning Model Training: Provides training data for predicting path delays, signal integrity, and other issues.

• Data Characteristics:

  1. Temporal Order: Elements in the array are strictly arranged in the chronological order of signal propagation.
  2. Alternating Structure: Nodes (node_*) and arcs (inst_arc_*/net_arc_*) appear alternately, forming a complete path.
  3. Incremental IDs: The IDs of nodes and arcs increase sequentially along the path, making it easy to trace the signal flow.
  4. Hierarchical Details: Net arcs contain nested, detailed wire segment information (edge_*), providing fine-grained interconnect characteristics.

Foundation Data: Graph

• Example File Path: vectors/wire_graph/timing_wire_graph.json

• Data Generation Mechanism:

  • Data Preparation: In the Vectorization::buildFeature method, _data_manager.buildLayoutData() is first called to build layout data, followed by _data_manager.buildGraphData() to construct the initial graph data structure.
  • Feature Extraction: The generateFeature() method is called to create a VecFeature object, and feature.buildFeatureTiming() generates timing-related features.
  • Timing Analysis: The runVecSTA method (called on an InitSTA::getInst() instance via eval_tp->runVecSTA()) performs a detailed timing analysis to calculate key parameters like arrival times and required times for nodes.
  • Data Saving: Finally, the _data_manager.saveData() method saves the generated Graph data as a JSON file.

• Data Structure Example: Graph data is organized in JSON format, using a standard graph data structure to represent the net connection relationships in the chip design.

  {
      "nodes": [ // Array of nodes, representing pins, ports, and routing nodes
        {
            "id": "node_0",
            "name": "clk",
            "is_pin": false,
            "is_port": true,
            "node_feature": {
                "is_input": true,
                "fanout_num": 1,
                "is_endpoint": false,
                "cell_name": "NA",
                "sizer_cells": [],
                "node_coord": [
                    0.0,
                    0.0
                ],
                "node_slews": [
                    0.0,
                    0.0,
                    0.0,
                    0.0
                ],
                "node_capacitances": [
                    0.0,
                    0.0,
                    0.0,
                    0.0
                ],
                "node_arrive_times": [
                    1.1e+20,
                    1.1e+20,
                    1.1e+20,
                    1.1e+20
                ],
                "node_required_times": [
                    1.1e+20,
                    1.1e+20,
                    1.1e+20,
                    1.1e+20
                ],
                "node_net_load_delays": [
                    0.0,
                    0.0,
                    0.0,
                    0.0
                ],
                "node_toggle": 1.0,
                "node_sp": 0.5,
                "node_internal_power": 0.0,
                "node_net_power": 2.8232123814158402e-05
            }
        },
        // more nodes...
      ],  
      "edges": [ // Array of edges, representing connections between nodes
        {
            "id": "edge_0",
            "from_node": 510,
            "to_node": 746,
            "is_net_edge": false,
            "edge_feature": {
                "edge_delay": [
                    0.088476,
                    0.051862,
                    0.088476,
                    0.051862
                ],
                "edge_resistance": 0.0,
                "inst_arc_internal_power": 0.00011665750115296025
            }
        },
        // more edges...
      ]   
  }
  

• Field Descriptions:

  • Nodes: Each node represents a connection point in the chip design, which could be a cell pin, a port, or a routing node.

    Field Name	Data Type	Description	Unit	Example Value
    id	string	Unique identifier for the node	-	"node_4"
    name	string	Node name, usually containing instance and pin names	-	"clk_0_buf:X"
    is_pin	boolean	Whether it is a cell pin	-	true
    is_port	boolean	Whether it is a chip port	-	false
    node_feature	object	Collection of node feature attributes, including timing and power info	-	-

  • Node Features (node_feature):

    Field Name	Data Type	Description	Unit	Example Value
    is_input	boolean	Whether it is an input node (relative to the cell)	-	false
    fanout_num	integer	Fanout count, indicating the number of loads driven by this node	-	17
    is_endpoint	boolean	Whether it is an endpoint of a timing path	-	false
    cell_name	string	Cell name, if the node is a cell pin	-	"sky130_fd_sc_hs__buf_1"
    sizer_cells	array[string]	List of replaceable cells for cell sizing optimization	-	["sky130_fd_sc_hs__buf_16", ...]
    node_coord	array[float]	Node coordinates (x, y)	DB Unit	[57.842, 61.732]
    node_slews	array[float]	Node slew times, includes values for four process corners	ns	[0.0, 0.0, 0.0, 0.0]
    node_capacitances	array[float]	Node capacitance values, includes values for four process corners	pF	[0.0, 0.0, 0.0, 0.0]
    node_arrive_times	array[float]	Signal arrival times at the node, includes values for four process corners	ns	[1.1e+20, 1.1e+20, ...]
    node_required_times	array[float]	Required arrival times at the node, includes values for four process corners	ns	[1.1e+20, 1.1e+20, ...]
    node_net_load_delays	array[float]	Node net load delays, includes values for four process corners	ns	[0.0, 0.0, 0.0, 0.0]
    node_toggle	float	Node toggle rate, indicating signal transitions per unit time	-	1.0
    node_sp	float	Signal probability, the probability of the signal being high	-	0.5
    node_internal_power	float	Node internal power, mainly cell internal power	W	0.0
    node_net_power	float	Node net power, mainly dynamic power on the net	W	0.00012901

  • Edges: Each edge represents a connection between two nodes, which can be an internal cell connection or a net connection.

    Field Name	Data Type	Description	Unit	Example Value
    id	string	Unique identifier for the edge	-	"edge_7960"
    from_node	integer	Start node ID (index in the nodes array)	-	7586
    to_node	integer	End node ID (index in the nodes array)	-	7587
    is_net_edge	boolean	Whether it is a net edge (true for net, false for internal cell connection)	-	true
    edge_feature	object	Collection of edge feature attributes, including delay and power info	-	-

  • Edge Features (edge_feature):

    Field Name	Data Type	Description	Unit	Example Value
    edge_delay	array[float]	Edge delay values, includes values for four process corners	ns	[0.0, 0.0, 0.0, 0.0]
    edge_resistance	float	Edge resistance value, mainly for net edges	Ohms	18.42857
    inst_arc_internal_power	float	Instance arc internal power, mainly for internal cell edges	W	0.0

• Data Usage:

  1. Timing Analysis: Provides a complete timing graph structure, supporting Static Timing Analysis (STA).
  2. Power Optimization: Contains detailed power-related parameters, supporting power analysis and optimization.
  3. Physical Design Verification: Offers the connection relationship between physical implementation and logical design, supporting design verification.
  4. Machine Learning Feature Extraction: Provides structured chip design feature data for ML models, supporting prediction and optimization tasks.
  5. Design Visualization: Supports converting the chip design into a visual graph structure for better understanding and analysis.

• Data Characteristics:

  1. STA Support: The Graph data includes complete timing-related information, supporting calculations and verifications related to Static Timing Analysis (STA).
  2. Process Corner Coverage: Key parameters (like delay, slew, etc.) include values for four process corners, supporting full process corner analysis.
  3. Hierarchical Structure: Data is organized by nodes and edges from graph theory, clearly expressing the connection relationships in the chip design.
  4. Multi-dimensional Features: Each node and edge contains multi-dimensional features like timing, power, and geometric location, supporting comprehensive analysis and optimization.

Foundation Data: Patch

• Example File Path: vectors/patchs/patch_0.json

• Data Generation Mechanism:

  1. The Vectorization::buildFeature function calls the buildPatchData method to construct the Patch data.
  2. The data generation process is managed by VecDataManager and specifically implemented by VecPatchDataManager.
  3. VecPatchInit is responsible for initializing the Patch grid and populating it with net and wire segment information.
  4. Finally, the data is saved as a JSON file via VecLayoutFileIO::saveJsonPatchs.

• Data Structure Example:

  {
      "id": 0,
      "patch_id_row": 0,
      "patch_id_col": 0,
      "llx": 0,
      "lly": 0,
      "urx": 4320,
      "ury": 4320,
      "row_min": 0,
      "row_max": 18,
      "col_min": 0,
      "col_max": 18,
      "cell_density": 0.8981481481481481,
      "pin_density": 40,
      "net_density": 0.12371399176954732,
      "macro_margin": 255180,
      "RUDY_congestion": 0.00029195559407049423,
      "EGR_congestion": 3.0,
      "timing": 0.0,
      "power": 3.2690654345212375e-07,
      "IR_drop": 0.0,
      "sub_nets": [
          {
              "id": 626,
              "llx": 1000,
              "lly": 0,
              "urx": 4320,
              "ury": 4320
          },
          // More sub-nets ...
      ],
      "patch_layer": [
          {
            "id": 4,
            "feature": {
                "wire_width": 140,
                "wire_len": 3120,
                "wire_density": 0.023405349794238684,
                "congestion": 2.0
            },
            "net_num": 1,
            "nets": [
              {
                "id": 650,
                "wire_num": 2,
                "wires": [
                    {
                        "id": 6600,
                        "feature": {
                            "wire_len": 3120
                        },
                        "path_num": 1,
                        "paths": [
                            {
                                "id1": 1151867,
                                "x1": 720,
                                "y1": 1200,
                                "real_x1": 720,
                                "real_y1": 1200,
                                "r1": 5,
                                "c1": 3,
                                "l1": 4,
                                "p1": -1,
                                "id2": 1158835,
                                "x2": 720,
                                "y2": 4320,
                                "real_x2": -1,
                                "real_y2": -1,
                                "r2": 18,
                                "c2": 3,
                                "l2": 4,
                                "p2": -1
                            }
                        ]
                    },
                    // More wire segments...
                  ]
              } 
            ]
          },
          // More patch_layers...
      ]
  }
  

• Field Descriptions:

  • Global Information

    Field Name	Data Type	Description	Unit	Example Value
    id	Integer	Unique identifier for the patch, as an integer sequence number	-	0
    patch_id_row	Integer	Row index of the patch in the grid	-	0
    patch_id_col	Integer	Column index of the patch in the grid	-	0
    llx	Integer	Lower-left x-coordinate of the patch	Database Unit	0
    lly	Integer	Lower-left y-coordinate of the patch	Database Unit	0
    urx	Integer	Upper-right x-coordinate of the patch	Database Unit	4320
    ury	Integer	Upper-right y-coordinate of the patch	Database Unit	4320
    row_min	Integer	Minimum row number of nodes within the patch	-	0
    row_max	Integer	Maximum row number of nodes within the patch	-	18
    col_min	Integer	Minimum column number of nodes within the patch	-	0
    col_max	Integer	Maximum column number of nodes within the patch	-	18
    cell_density	Float	Cell density within the patch, value range [0, 1]	-	0.898148
    pin_density	Float	Pin density within the patch	Number of pins	40
    net_density	Float	Net density within the patch	-	0.123714
    macro_margin	Integer	Maximum margin distance to a macro cell within the patch	Database Unit	255180
    RUDY_congestion	Float	Congestion estimation value based on the RUDY model	-	0.000292
    EGR_congestion	Float	Congestion estimation value based on the EGR model	-	3.0
    timing	Float	Timing information	ns	0.0
    power	Float	Power information	W	3.26907e-07
    IR_drop	Float	IR drop information	V	0.0
    sub_nets	Array	Information about sub-nets contained within the patch	-	[Array of sub-net objects]
    patch_layer	Array	Detailed information for each layer of the patch	-	[Array of layer objects]

  • Sub-net Information (sub_nets): Each element in the sub_nets array is a sub-net, defining the overlapping area of a net with the patch, as follows:

    Field Name	Data Type	Description	Unit	Example Value
    id	Integer	Unique identifier for the sub-net	-	626
    llx	Integer	Lower-left x-coordinate of the sub-net	Database Unit	1000
    lly	Integer	Lower-left y-coordinate of the sub-net	Database Unit	0
    urx	Integer	Upper-right x-coordinate of the sub-net	Database Unit	4320
    ury	Integer	Upper-right y-coordinate of the sub-net	Database Unit	4320

  • Layer Information (patch_layer): Each element in the patch_layer array represents a layer within the patch, defined as follows:

    Field Name	Data Type	Description	Unit	Example Value
    id	Integer	Layer ID, corresponding to a metal layer in the physical design	-	4
    feature	Object	Layer feature information	-	See table below
    net_num	Integer	Number of nets contained in the layer	-	1
    nets	Array	List of nets in the layer	-	[Array of net objects]

  • Layer Features (feature)

    Field Name	Data Type	Description	Unit	Example Value
    wire_width	Integer	Wire segment width	Database Unit	140
    wire_len	Integer	Total length of wire segments	Database Unit	3120
    wire_density	Float	Wire density	-	0.023405
    congestion	Float	Congestion level of this layer	-	2.0

  • Net Information (nets): Each element in the nets array represents a net, defined as follows:

    Field Name	Data Type	Description	Unit	Example Value
    id	Integer	Unique identifier for the net	-	650
    wire_num	Integer	Number of wires in the net	-	2
    wires	Array	List of wires	-	[Array of wire objects]

  • Wire Information (wires): Each element in the wires array represents a wire segment, defined as follows:

    Field Name	Data Type	Description	Unit	Example Value
    id	Integer	Unique identifier for the wire	-	6600
    feature	Object	Wire feature information	-	{"wire_len": 3120}
    path_num	Integer	Number of paths contained in the wire	-	1
    paths	Array	List of paths	-	[Array of path objects]

  • Path Information (paths): Each element in the paths array represents a segment of a wire's path, defined as follows:

    Field Name	Data Type	Description	Unit	Example Value
    id1	Integer	Unique identifier for the start point	-	1151867
    x1, y1	Integer	Grid coordinates of the start point	Database Unit	720, 1200
    real_x1, real_y1	Integer	Actual physical coordinates of the start point	Database Unit	720, 1200
    r1, c1	Integer	Row and column of the grid where the start point is located	-	5, 3
    l1	Integer	Layer ID where the start point is located	-	4
    p1	Integer	Pin index of the start point, -1 for non-real pins	-	-1
    id2	Integer	Unique identifier for the end point	-	1158835
    x2, y2	Integer	Grid coordinates of the end point	Database Unit	720, 4320
    real_x2, real_y2	Integer	Actual physical coordinates of the end point	Database Unit	720, 4320
    r2, c2	Integer	Row and column of the grid where the end point is located	-	18, 3
    l2	Integer	Layer ID where the end point is located	-	4
    p2	Integer	Pin index of the end point, -1 for non-real pins	-	-1
    via	Integer	Via ID (optional, exists only for cross-layer connections)	-	10 (Via number)

• Data Usage:

  1. Physical Layout Analysis: Provides detailed physical characteristics of different regions in a chip design, supporting designers in macroscopic layout evaluation.
  2. Layout Visualization: The data structure facilitates the graphical display and analysis of layout regions.
  3. Performance Optimization Guidance: Based on path information and electrical characteristics, it provides detailed physical foundation data for timing performance optimization.
  4. AI Model Training Data: Provides structured physical design feature data for deep learning models, supporting AI tasks like routing prediction and congestion analysis.

• Data Characteristics:

  1. Patch data is divided into grid-like patch units based on physical design regions.
  2. Each patch contains information for multiple layers, and each layer includes detailed geometric and electrical attributes of nets, wires, and paths.
  3. The data is organized hierarchically from macro to micro: patch -> layer -> net -> wire -> path.
  4. Path information precisely describes the connection points on a wire, including coordinates, layer, and grid position.

Foundation Data: Instance

• Example File Path: vectors/instances/instances.json

• Data Generation Mechanism:

  1. The Vectorization::buildFeature function calls the _data_manager.saveData method to save all data.
  2. The saveData method internally calls the VecLayoutFileIO::saveJson function.
  3. The saveJson function calls the saveJsonInstances method specifically to generate Instance data.
  4. Finally, the data is saved in JSON format to the specified directory.

• Data Structure Example: The instances.json file uses the following structure:

  {
    "instance_num": 1448, // Total number of instances
    "instances": [
      {
        "id":0,
        "cell_id":236,
        "name":"ctrl/_17_",
        "cx":78480,
        "cy":61605,
        "width":1440,
        "height":3330,
        "llx":77760,
        "lly":59940,
        "urx":79200,
        "ury":63270
        },
      // More instances...
    ]
  }
  

• Field Descriptions:

  The fields for each instance are described as follows:

  Field Name	Data Type	Description	Unit	Example Value
  id	Integer	Unique identifier for the instance	-	1
  cell_id	Integer	ID of the cell master to which the instance belongs	-	1001
  name	String	Instance name	-	"sky130_fd_sc_hs__and2_1_0"
  cx	Float	x-coordinate of the instance's center point	Database Unit	1234.5
  cy	Float	y-coordinate of the instance's center point	Database Unit	5678.9
  width	Float	Instance width	Database Unit	420.0
  height	Float	Instance height	Database Unit	180.0
  llx	Float	Lower-left x-coordinate of the instance	Database Unit	1024.5
  lly	Float	Lower-left y-coordinate of the instance	Database Unit	5588.9
  urx	Float	Upper-right x-coordinate of the instance	Database Unit	1444.5
  ury	Float	Upper-right y-coordinate of the instance	Database Unit	5768.9

• Data Usage:

  1. Physical Layout Representation: The instance data precisely describes the physical location and size of each standard cell in the design.
  2. Timing Analysis Support: The location information of instances is fundamental for calculating wire lengths and interconnect delays.
  3. Routability Analysis: The distribution of instances is used to evaluate layout quality and routing difficulty.
  4. AI Model Training: Provides real physical design instance layout data for machine learning models.

• Data Characteristics:

  1. Comprehensiveness: Contains information for all instances in the design, with no omissions.
  2. Precision: Coordinate and size information is accurate to the database unit.
  3. Associativity: Linked to cell library information via cell_id.
  4. Uniform Structure: All instances use the same data structure, facilitating batch processing.
  5. Extensibility: The JSON format supports the addition of extra instance attributes.

Foundation Data: Instance Graph

• Example File Path: vectors/instance_graph/timing_instance_graph.json

• Data Generation Mechanism:

  1. The Vectorization::buildFeature function calls the generateFeature method.
  2. The generateFeature method internally calls the VecFeature::buildFeatureTiming function.
  3. The buildFeatureTiming function calls eval_tp->getTimingInstanceGraph() to obtain the instance graph data.
  4. Finally, the data is saved in JSON format to the specified directory via the SaveTimingInstanceGraph function.

• Data Structure Example:

  The timing_instance_graph.json file uses the following structure:

  {
    "nodes": [        
      {
        "id": "node_0",
        "name": "ctrl/_34_",
        "leakage_power": 1.6979909062047418e-10
      },
      // More nodes...
    ],
    "edges": [
      {
        "id": "edge_0",
        "from_node": 0,
        "to_node": 1
      },
      // More edges...
    ]
  }
  

• Field Descriptions:

  • Nodes: Each node object represents an instance and contains the following fields:

    Field Name	Data Type	Description	Unit	Example Value
    id	String	Unique identifier for the node, format "node_number"	-	"node_0"
    name	String	Instance name	-	"ctrl/34"
    leakage_power	Float	Leakage power of the instance	W	1.697e-10

  • Edges: Each edge represents the connection relationship between two instances and contains the following fields:

    Field Name	Data Type	Description	Unit	Example Value
    id	String	Unique identifier for the edge, format "edge_number"	-	"edge_1037"
    from_node	Integer	Index of the driving node (corresponds to the index in the nodes array)	-	640
    to_node	Integer	Index of the receiving node (corresponds to the index in the nodes array)	-	188

• Data Usage:

  1. Timing Analysis: The instance graph provides all instances and their connection relationships in the design, which can be used for timing path analysis and delay prediction.
  2. Power Estimation: Contains leakage power information, which can be used for overall power estimation.
  3. Physical Constraint Study: Reflects the connection density between instances, which helps in studying the impact of physical constraints on the design.
  4. AI Model Training: Provides instance-level connection relationship and power data for machine learning models.

• Data Characteristics:

  1. Structured Representation: Uses a graph data structure to clearly represent the connection relationships between instances.
  2. Lightweight Design: Edges contain only basic connection information, facilitating fast processing and analysis.
  3. Includes Power Information: Each node includes leakage power data, which is convenient for power analysis.
  4. Global View: Provides a global view of the instance connections for the entire design.

Foundation Data: Cell

• Example File Path: vectors/tech/cells.json

• Data Generation Mechanism:

  1. The Vectorization::buildFeature function calls _data_manager.buildLayoutData to construct the basic layout data.
  2. After data construction is complete, _data_manager.saveData is called to save the data.
  3. During the saving process, the VecLayoutFileIO::saveJson method calls the saveJsonCells method.
  4. The saveJsonCells method retrieves cell data from _layout->get_cells().get_cell_map() and saves it to the cells.json file.

• Data Structure Example:

  {
    "cell_num": integer,      // Total number of cell_master units
    "cells": [            // Array of cells
      {
        "id": integer,       // Cell ID
        "name": "string", // Cell name
        "width": integer,    // Cell width
        "height": integer    // Cell height
      },
      // More cell masters...
    ]
  }
  

• Field Descriptions:

  Field Name	Type	Description	Unit	Example Value
  cell_num	integer	Total number of cells in the cell library	-	416
  cells	array	Array containing information for all cells	-	[{}, ...]
  id	integer	Unique identifier for the cell	-	415
  name	string	Cell name, including technology information and functional description	-	"sky130_sram_1rw1r_44x64_8"
  width	integer	Physical width of the cell	Database Unit (e.g., nm)	445240
  height	integer	Physical height of the cell	Database Unit (e.g., nm)	237415

• Data Usage:

  1. Physical Design Reference: Provides physical dimension information for standard cells and macro cells for placement and routing.
  2. Area Estimation: Used for chip area calculation and density analysis.
  3. Power Analysis: Serves as base data for power calculation.
  4. Timing Optimization: Provides cell-level references for timing analysis.

• Data Characteristics:

  1. Completeness: Contains information on all available standard cells and macro cells.
  2. Physical Accuracy: Provides precise physical dimension parameters.
  3. Process-Related: The data reflects cell characteristics under a specific technology node.
  4. Structured Storage: Uses JSON format for easy parsing and processing by programs.
  5. Extensibility: The data structure is simple and clear, making it easy to add new cell attributes.

Foundation Data: Technology

• Example File Path: vectors/tech/tech.json

• Data Generation Mechanism:

  1. The Vectorization::buildFeature function calls _data_manager.buildLayoutData to construct the basic layout data.
  2. After data construction is complete, _data_manager.saveData is called to save the data.
  3. During the saving process, the VecLayoutFileIO::saveJson method calls the saveJsonTech method.
  4. The saveJsonTech method retrieves layer data from _layout->get_layout_layers().get_layout_layer_map() and via data from _layout->get_via_name_map(), saving them to the tech.json file.

• Data Structure Example:

  {
    "layer_num": integer,  // Number of technology layers
    "layers": [        // Array of layer information
      {
        "id": integer,   // Layer number
        "name": "string" // Layer name
      },
      // More layers...
    ],
    "via_num": integer,   // Number of via types
    "vias": [          // Array of via information
      {
        "id": integer,   // Via number
        "name": "string", // Via name
        "bottom": {   // Bottom metal enclosure area coordinates
          "llx": integer,
          "lly": integer,
          "urx": integer,
          "ury": integer
        },
        "cut": {      // Via cut area coordinates
          "llx": integer,
          "lly": integer,
          "urx": integer,
          "ury": integer
        },
        "top": {      // Top metal enclosure area coordinates
          "llx": integer,
          "lly": integer,
          "urx": integer,
          "ury": integer
        },
        "row": integer,  // Number of via rows
        "col": integer,  // Number of via columns
        "bottom_direction": "string", // Offset direction of the cut relative to the bottom enclosure
        "top_direction": "string"     // Offset direction of the cut relative to the top enclosure
      },
      // More vias...
    ]
  }
  

• Field Descriptions:

  • Layer Information Fields

    Field Name	Type	Description	Unit	Example Value
    layer_num	integer	Total number of layers in the technology	-	11
    layers	array	Array of detailed information for all layers	-	[{}, ...]
    id	integer	Sequential number of the layer (from bottom to top)	-	10
    name	string	Layer name (metal layer, via layer, etc.)	-	"met5"

  • Via Information Fields

    Field Name	Type	Description	Unit	Example Value
    via_num	integer	Total number of via types	-	32
    vias	array	Array of detailed information for all via types	-	[{}, ...]
    id	integer	Unique identifier for the via type	-	0
    name	string	Via type name	-	"L1M1_PR"
    bottom	object	Bounding box coordinates of the bottom metal enclosure area	Database Unit	{"llx": -85, "lly": -85, "urx": 85, "ury": 85}
    cut	object	Bounding box coordinates of the via cut area	Database Unit	{"llx": -85, "lly": -85, "urx": 85, "ury": 85}
    top	object	Bounding box coordinates of the top metal enclosure area	Database Unit	{"llx": -145, "lly": -115, "urx": 145, "ury": 115}
    row	integer	Number of rows of via cuts	-	1
    col	integer	Number of columns of via cuts	-	1
    bottom_direction	string	Connection direction for the bottom layer (C-Center, N-North, S-South, E-East, W-West)	-	"C"
    top_direction	string	Connection direction for the top layer (C-Center, N-North, S-South, E-East, W-West)	-	"C"

• Data Usage:

  1. Physical Design Rule Checking: Provides detailed parameters for layers and vias to verify if the design complies with technology rules.
  2. Routing Guidance: Serves as technology constraint input for automatic routing tools.
  3. Design Visualization: Provides accurate layer and via representations for design visualization.
  4. Timing Analysis: Provides necessary technology parameters for signal integrity and timing analysis.

• Data Characteristics:

  1. Technology Accuracy: Accurately reflects layer and via characteristics for a specific technology node.
  2. Completeness: Includes all necessary physical layers and via types.
  3. Structured Representation: Uses JSON format for easy parsing and processing by programs.
  4. Complete Spatial Information: Provides precise geometric coordinates for different parts of each via.
  5. Directional Information: Includes directional information for via connections, which helps in understanding routing topology.

Data Owner

iEDA Team

License

GPL