168 Four-MAC Core Neural Network Inference Engine

168 : Four-MAC Core Neural Network Inference Engine

Design render

How it works

Four signed 4-bit MAC lanes compute four dot products per job, with 20-bit accumulation. A shared activation and four weights arrive as three acknowledged bytes per vector element. The host supplies START (or START_BIASED followed by four signed 20-bit biases), streams a nonempty vector with LAST on its final element, and reads twelve result bytes. Models with more than four output neurons use successive jobs and replay the activations.

The current top module returns raw signed sums including optional bias. Activation integration, requantization, and multilayer scheduling are performed by the host or remain future work. A standalone ReLU block exists but is not connected to this datapath. The 50 MHz clock is a design target pending SKY130 physical timing verification.

How to test

Install test/requirements.txt in a Python virtual environment and run make check from the repository root. This runs lint, functional tests, block simulations, and pin-level image/protocol regressions. See docs/input_protocol.md for the exact byte format, handshakes, and clock-domain stability requirements.

On hardware, configure the host to drive ui[7:0], uio[0], uio[2], and uio[3] only; all other uio pins are status outputs. Hold RX_REQ and TX_ACK low during reset. After reset release and JOB_READY, send command 01 followed by operand packets, or command 03 followed by twelve bias bytes and then operand packets. Acknowledge each output byte individually and check PROTOCOL_ERROR. The status and data handshakes permit a slow host, but the project clock must remain running.

External hardware

Tiny Tapeout demo board or a compatible external host supplying the project clock, reset, and acknowledged byte I/O. Host memory holds images, weights, and intermediate layer results. See host/input_protocol.py for preprocessing and packet helpers. The SKY 26c digital development configuration does not yet implement the requested analog-slot physical floorplan; see docs/sky130_setup.md before submission.

IO

#InputOutputBidirectional
0RX_DATA[0]TX_DATA[0]RX_REQ
1RX_DATA[1]TX_DATA[1]RX_ACK
2RX_DATA[2]TX_DATA[2]RX_COMMAND
3RX_DATA[3]TX_DATA[3]TX_ACK
4RX_DATA[4]TX_DATA[4]TX_VALID
5RX_DATA[5]TX_DATA[5]TX_LAST
6RX_DATA[6]TX_DATA[6]JOB_READY
7RX_DATA[7]TX_DATA[7]PROTOCOL_ERROR

Chip location

Controller Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Analog Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux Analog Mux Mux Mux Mux Mux Mux Mux Mux Mux Mux tt_um_chip_rom (Chip ROM) tt_um_factory_test (Tiny Tapeout Factory Test) tt_um_teuscher_eml_fabric (EML Fabric — analog exp/ln compute cells) tt_um_wokwi_465656663515438081 (Convert binary to hex on 7 segments) tt_um_nikita_face_detect (FPGA Face Detection) tt_um_obstacle_avoider (Obstacle Avoider State Machine) tt_um_poket_animal (Poket Animal) tt_um_drewbabel_uart (Configurable FIFO-buffered UART with APB CSR) tt_um_wokwi_469163916296039425 (TT Workshop Test) tt_um_jonahsaunders_slsvga (tt_um_jonahsaunders_slsvga) tt_um_fatigue_monitor (Fatigue Monitor (PPG Pulse-Interval Variability)) tt_um_vedam_dual_port_ram (Dual Port RAM) tt_um_wokwi_469739097665887233 (Tiny Tapeout Template Copy) tt_um_spdif_to_i2s_kilpelaj (S/PDIF to I2S receiver) tt_um_morse_converter (ASCII to Morse Code Converter) tt_um_wokwi_469806914724000769 (Spin, Text and VGA) tt_um_wokwi_469701770572338177 (TinyTapeout) tt_um_garnetkoebel_communotron (Communotron) tt_um_wokwi_469449970323169281 (full adder) tt_um_duzabf_2026_ow (A WIP Online Workshop 2026 project) tt_um_wokwi_469807513638180865 (Tiny Tapeout NAK) tt_um_ttsky26c_oguz (ttsky26c-202607-mehmetoguzderin by Oguz) tt_um_kashif_fp4_sparse_tpu (FP4 Sparse Mini-TPU) tt_um_moein_maleki_arm16 (arm16) tt_um_wokwi_469453454643027969 (ON Check System) tt_um_felixcheng_neural_core (Neural Compute Core (V0.15)) tt_um_wokwi_469788774011248641 (Spin Display - select-reset-reverse) tt_um_wokwi_469449443070765057 (Samuel's first chip) tt_um_wokwi_469449007236383745 (testinttrsv01) tt_um_vga_ca (VGA cellular Automaton) tt_um_dosci_500hz (Digital Oscillator 500 Hz) tt_um_wokwi_469747443569078273 (XOR test project - Tiny Tapeout workshop) tt_um_wokwi_469585758593419265 (spinner) tt_um_fp16_mac (FP32 Math Unit) tt_um_1DC_vga_dyoa (VGA Design Your Own ASIC) tt_um_haydenevans_top (Systolic Processing Element) tt_um_wokwi_469806252715961345 (TT_Proj_SA) tt_um_wokwi_469448996577604609 (Tiny Tapeout - Reto) tt_um_ehofmannbr_pmodvga_06 (VGA Color Tiles) tt_um_lfglabs_lsc1u (leanSilicon LSC-1 Micro arithmetic kernel) tt_um_wokwi_469804280240495617 (Zetterling SRAM) tt_um_wokwi_469806066852696065 (TileTestchase) tt_um_wokwi_469449118072978433 (binary_add_v1) tt_um_voltage_amplifier_neuron (Voltage Amplfier Neuron) tt_um_wokwi_469449686545956865 (Tiny Tapeout Template Copy_JinoShiono) tt_um_wokwi_469448887171240961 (Tiny Tapeout - Mini CORDIC) tt_um_wokwi_469809033878555649 (Tiny Tapeout Yummy Chip - bgianfo) tt_um_sirajmuhammad_bpsk_mod (BPSK Baseband Modulator) tt_um_K_coder_9 (TENs device frequency controller) tt_um_wokwi_469758119198926849 (LL_6BitShiftRegister_ToggleEnabledFeedback) tt_um_Asaadkhex_6x6u (6x6 UART Bussbar Switch) tt_um_wokwi_469809198944364545 (tt8-8bit-cpu Copy) tt_um_wokwi_469710279607305217 (Tiny Tapeout Submission KL - SiliDize) tt_um_wokwi_469629799092815873 (2:1 Mux with differential outputs) tt_um_poundbrad_reciprocal_counter (Two-Channel Reciprocal Counter) tt_um_joonatanalanampa_cordic (CORDIC-1) tt_um_x4ntha_nova (Data General Nova 1200 CPU) tt_um_quick_bus (quick_bus) tt_um_wokwi_470058539448408065 (Nigel's Tiny Tapeout Project) tt_um_wokwi_470058244557293569 (Tiny Tapeout Kabisan) tt_um_wokwi_470058241869790209 (Abdi's desgin) tt_um_wokwi_470060107756808193 (Sukhraj Deol's Chip) tt_um_wokwi_470058578588614657 (The Chip of Master George Stead) tt_um_wokwi_470069286344622081 (Tiny Tapeout ISHA) tt_um_ucl_display (Flashing... lights) tt_um_wokwi_470058746279043073 (Arihant's first Wokwi design) tt_um_wokwi_470060103260512257 (Tiny Tapeout Jabriel Copy) tt_um_wokwi_470069460157662209 (haadi's tiny tapeout) tt_um_wokwi_470058418706939905 (Kitty) tt_um_wokwi_470058490118136833 (Iris) tt_um_wokwi_470060098828179457 (Temz_ tiny tapeout) tt_um_wokwi_470058023187099649 (Osman WOKWI project 1) tt_um_wokwi_470057988621827073 (Viraj Tiny Template Full Adder TEST) tt_um_wokwi_470069802034377729 (Tiny Tapeout Template Copy) tt_um_wokwi_470070136685362177 (full adder) tt_um_wokwi_470070449402211329 (Anastasia Copy (2)) tt_um_wokwi_470059864883484673 (Keyaan’s first Wokwi design) tt_um_wokwi_470071200164912129 (full adder tiny tapeout Copy) tt_um_wokwi_470060671178857473 (SBUSixth First Chip Design Mentored by Tiny Tapeout) tt_um_wokwi_470099562753182721 (Isaac Tiny Tapeout) tt_um_wokwi_470120538476737537 (efwz8voices) tt_um_lelo_gr01_analogicus (LELO-GR01) tt_um_lelo_gr04_analogicus (LELO-GR04) tt_um_lelo_gr02_analogicus (LELO-GR02) tt_um_pump_out (60 Hz RMS Pump-Out Controller) tt_um_urish_simon (Simon Says memory game) tt_um_lelo_gr03_analogicus (LELO-GR03) tt_um_wokwi_470299374901578753 (Shrimp) tt_um_vga_clock (VGA clock) tt_um_frequency_counter (Frequency counter) tt_um_z2a_rgb_mixer (RGB Mixer demo) tt_um_mattvenn_r2r_dac_3v3 (Analog 8 bit 3.3v R2R DAC) tt_um_rebeccargb_universal_decoder (Universal Binary to Segment Decoder) tt_um_rebeccargb_hardware_utf8 (Hardware UTF Encoder/Decoder) tt_um_rebeccargb_intercal_alu (INTERCAL ALU) tt_um_rebeccargb_vga_pride (VGA Pride) tt_um_ogggggish_ota_ldo (SSF Capless LDO) tt_um_hariri4534_audioplayback (audioplayback) tt_um_wokwi_470637150792846337 (Joni - Tiny Tapeout Teardown2026 Workshop) tt_um_wokwi_470635013242210305 (Tom's first Wokwi design) tt_um_wokwi_470635780983408641 (Tiny Tapeout-AyeshaTeardown26) tt_um_wokwi_470639152626282497 (KeKoaM Tiny Tapeout) tt_um_wokwi_470637073520124929 (Tiny Tapeout workshop) tt_um_toby43479_iox (IO Expander with PWM) tt_um_wokwi_470635764113915905 (Divider Demo) tt_um_wokwi_470635580461052929 (Mann-teardown-project) tt_um_wokwi_470639672984256513 (KCs 001 TinyTapeout Design) tt_um_wokwi_470635507665754113 (Tiny Tapeout Template Copy) tt_um_wokwi_470637047364443137 (Pixel-Curio-Chip) tt_um_terihear_tinytearout (TinyTearout) tt_um_wokwi_470643025042834433 (TT 2026) tt_um_wokwi_470637360757626881 (Tiny Tapeout Template Copy) tt_um_wokwi_470635627278929921 (Tiny Tapeout Workshop) tt_um_wokwi_474471160110403585 (Cylon-Scanner) tt_um_wokwi_470646659230201857 (bloopbloop) tt_um_pthomas_sigma_delta (Continuous-Time Sigma-Delta ADC (1st order)) tt_um_sky_tpu_3x3 (Sky TPU 3x3) tt_um_tpcannon7_fir (tinyfir) tt_um_bruniliomuy_top (Fir_Filter) tt_um_semiqa_diff_opamp (Diff-In-Diff-Out-OpAmp) tt_um_TinyProcessor_naiyar_ (TinyProcessor) tt_um_CCDmos3D (ADC for CCDmos3D pixel) tt_um_snn_lif_neuron (snn_lif_neurons) tt_um_galaguna_NanoSys_fit (Nano-120_CPU@ler.uam.mx) tt_um_rowles_regime (Single-Bit Macro Regime Classifier) tt_um_rowles_fedmodel (The Fed Model (F1/F2)) tt_um_sky26c (tt_sky26c) tt_um_aka_regfile_ecc (regfile_ecc) tt_um_fwilson12_mac (int8 MAC) tt_um_davidbroughsmyth_ecg_sar12 (heart_monitor_adc_art) tt_um_foxworks_picorv32 (TCD Foxworks PicoRV32) tt_um_saltworks_ndf_c32 (Neural dataflow fabric — bit-serial MAC cells on a self-routing switch) tt_um_yjeum11 (DTMF (Touch-Tone) decoder) tt_um_vedic_mult (4-bit Vedic Multiplier) tt_um_atx_phased_interferometer (Acoustic Interferometer) tt_um_tilesos_dual_adc (Dual-Path Noise-Shaping ADC) tt_um_darga_cirom (Darga CiROM digital read + ternary MAC) tt_um_azara_cirom (Azara CiROM ternary read) tt_um_spi_reg_bank (8-bit Modified RISC-V) tt_um_aialaqili_updown_counter (4-bit Up/Down Counter) tt_um_noahzperez29_riscv_core (Noah RISC-V Core) tt_um_fp8_fpu (FP8 (E4M3) Floating-Point Unit) tt_um_costinemanuelv_gps_daily_trigger (GPS Daily Trigger) tt_um_ja_achtung_1x1 (JA Achtung Compact) tt_um_ja_achtung_1x2 (JA Achtung Full) tt_um_pwm_spice (spice-pwm-tapeout) tt_um_wecallemjazzyfact_bgr_ldo (BGR + LDO 3.3V/1.8V Integrated IP) tt_um_lelo_temp_wulffern (LELO-TEMP) tt_um_wokwi_472389622799861761 (3-Bit 101 Pattern Detector) tt_um_LnL_SoC (Lab and Lectures SoC) tt_um_dash_lucas_risc (risc_processor) tt_um_serdes_ephotonics (UCIe-style SERDES with analog TX driver & RX slicer) tt_um_joram200 (Kalman Filter Hardware Accelerator) tt_um_colbywonn_poly_synth (Poly Synth v1.0) tt_um_nobleg30_uart_vga_scroller (UART VGA Text Scroller) tt_um_multi_precision_mult (Multi-Precision Multiplier) tt_um_pratibha_munnangi_qkt_mac (QKT MAC Accelerator) tt_um_akankaan_bf16_fma (BF16 Fused Multiply-Add (FMA)) tt_um_rtfce (RTFCE - Reconfigurable Temporal Fault/Constraint Engine) tt_um_hdc_classifier (HDC Classifier) tt_um_preethi8a_adaptive_lfsr_prng (Self-Seeding Adaptive 16-bit Galois LFSR PRNG) tt_um_dilip951_cpu_systolic_array (Reconfigurable mixed-precision 2x2 systolic MAC array) tt_um_pqc_ntt_bfly (Crypto-Agile NTT Butterfly (ML-KEM / ML-DSA / FN-DSA)) tt_um_mlkem_coefficient_integrity (Fault-Aware Constant-Time FO Backend for ML-KEM) tt_um_vital_ap (VITAL-AP: Adaptive Pixel Register) tt_um_olaf8 (OLAF-8: Bounded-Memory Online Adaptive Fuzzy Inference) tt_um_Median_MAD (Streaming Median-MAD Estimator) tt_um_tnt_mosbius (tnt's variant of SKY130 mini-MOSbius) tt_um_undip_ann_q610 (UNDIP ANN Accelerator (SPI + bring-up self-test)) tt_um_cpu8 (CPU8) tt_um_vaishnavipatil5_configurable_cam (Configurable CAM with Masked Pattern Matching and Priority Resolution) tt_um_gina_env_monitor (Environmental Mapping Processor) tt_um_manasvibhat_bloom_filter (Bloom Filter Membership Tester) tt_um_amazing_sage_snn (LIF Neuron SNN) tt_um_nkanderson_lut_snn (LUT Spiking Network Classifier) tt_um_bigmanraffa_clm (Clementine: 4-lane int8 SIMT GPU) tt_um_adityarprasad_fft (Adaptive-Precision FFT) tt_um_oscillating_bones (Oscillating Bones) tt_um_silicon_edge_ns_sar_adc (NS SAR ADC) tt_um_sishi888_tinymind (TinyMind SoC) tt_um_afra_123_ecc_memory (Runtime-Reconfigurable ECC Memory) tt_um_kenchangh_mnist (MNIST Digit Recognition) tt_um_ece298a_8_bit_cpu_top (8-Bit CPU) tt_um_libormiller_SIMON_V2 (SIMON V2) tt_um_WaiMingLee888_nanov_1tile (NanoV RV32E one-tile RISC-V processor) tt_um_four_bit_nn_accel (4-bit Neural Network Accelerator) tt_um_rsa_simple (RSA Simple Encryptor) tt_um_synapticrw_lif_neuron (LIF Neuron (SynapticRW Teardown 2026)) tt_um_smunigan_ipv4_filter (IPv4 Header Filter) tt_um_jjy_spi_watchdog (SPI-Configurable Watchdog Timer) tt_um_osian_beam_controller (Programmable Metasurface Beam Controller) tt_um_namramazhar_popcnt_shiftreg (17-bit Wallace-tree POPCNT with shift-register input) tt_um_obookstay_puf (An arbiter PUF) tt_um_arminkardovic_montenegro_securekey (Montenegro SecureKey) tt_um_rcyaon_droop (All-Digital Supply Droop Detector) tt_um_ctw_spms (CTW-SPMS — Programmable Smart Power Management & Supervisor) tt_um_taiwoopesade_tempo_detector_sky26c (Hardware Audio Tempo Detector) tt_um_wokwi_470059878406973441 (Ehan's first TinyTapeout Project) tt_um_wokwi_470637170309995521 (My First Wokwi Thing!) tt_um_wokwi_470637401137246209 (Teardown Tiny Tapeout) tt_um_wokwi_469443433165025281 (Tiny Tapeout First Design Beth Plummer) tt_um_wokwi_472423526521678849 (4-bit to 5x7 Matrix Decoder for Tiny Tapeout) tt_um_wokwi_470057961258181633 (Tiny Tapeout Template Kavana) tt_um_wokwi_470057993933917185 (ivane- Tiny Tapeout (full adder)) tt_um_wokwi_470088776251343873 (training_project_kaylem) tt_um_neuropong (NeuroPong) tt_um_tamagotchi (TamaGotThis) tt_um_group02_seethebeat (SeeTheBeat) tt_um_kul_chromechain (Chrome Chain) tt_um_baked_weights (Baked-Weights Shakespeare GPT) tt_um_gilangfajrul_sar_adc (sar-adc) tt_um_Logy_FMAC (FMAC) tt_um_porkfreezer_rrio_opamp (RRIO Op-amp) tt_um_diff_engine (DSLX finite_difference) tt_um_dragonochi (WISH) tt_um_siliconsonics (ultrasonic sonar: range and bearing) tt_um_kul_conway (Interactive Conway's Game of Life) tt_um_algofoogle_ttsky26c_analog (Assorted analog in 1 tile) tt_um_mariavictoriaalm_qubit_sim ( tt-2qubit-sim) tt_um_andre_dpe (Dot product engine) tt_um_rmranjitkarNULL_pong_top (last_minute_Pong) tt_um_SAR_ADC (CTW LDO and Dynamic Comparator) tt_um_fabulous_sky_26c (Tiny FABulous FPGA) tt_um_tomvdsch_tiny32_soc (Tiny32 RV32IMA Zephyr-target SoC) tt_um_np523_pong (Pong) tt_um_usfq_adc_procmon (USFQ 8-bit Tracking ADC and Process Variation Monitor) tt_um_rangfuu_alu (Tiny ALU PD) tt_um_wokwi_473800139156677633 (Tiny Snake with PRISM 8) tt_um_mini_nn (Four-MAC Core Neural Network Inference Engine) tt_um_kianv_rv32_regfile (KianV uLinux RISC-V regfile edition) tt_um_2048_vga_game (2048 sliding tile puzzle game (VGA)) tt_um_urish_rings (VGA Rings) tt_um_silicon_art_vga_screensaver (VGA Screensaver with Silicon Art ROM) tt_um_rom_vga_screensaver (VGA Screensaver with embedded bitmap ROM) tt_um_krisjdev_manchester_baby (Manchester Baby) tt_um_urish_sic1 (SIC-1 8-bit SUBLEQ Single Instruction Computer) tt_um_ThomasCowieEngineering_LMC (Little Man Computer CPU) tt_um_pranavUl_ascon_aead128 (Ascon bit-serial permutation engine) tt_um_orca (ORCA — Online Reconfigurable Circuit with Adaptation) tt_um_krisjdev_artwork (Silicon Artwork) tt_um_htfab_caterpillar (Simon's Caterpillar) tt_um_htfab_vga_tester (Video mode tester) Available Available Available Available Available Available Available Available Available Available